Cheng Liu

Developing intelligent fault diagnosis (IFD) methods based on deep neural networks has attracted considerable attention and brought successful breakthroughs for industrial applications in recent years. However, the historical IFD methods often concentrate on single diagnosis tasks, such as emerging fault detection or compound fault decoupling (CFD), leading to potential misdiagnosis and missed diagnosis, and also necessitate sufficient labeled data for model training in advance. Aiming to address such problems, a collaborative fault diagnosis framework, named the Open-Set Capsule Network (OpensetCapsNet), is proposed to mitigate the misdiagnosis of both emerging and compound faults. First, the OpensetCapsNet is constructed with four parts: a feature extractor for feature learning, an emerging fault detector for unknown fault detection, a compound fault classifier for intelligent CFD, and a domain discriminator for domain adaptation. Second, a collaborative training strategy is proposed based on adversarial learning and multitask learning. This strategy enables training the OpensetCapsNet with datasets that include both health and known fault samples collected from one working condition, alongside unlabeled samples from varying conditions. Finally, the cross-validation experiments were conducted on an automobile transmission, demonstrating the OpensetCapsNet’s capacity for collaborative execution of diagnosis tasks, encompassing known fault classification, emerging fault detection, and CFD. The proposed framework significantly reduces the risk of misdiagnosis and missed diagnosis, offering a substantial advancement for IFD in industrial applications.

Industrial robotic arms in modern industry face rising cyber-physical attacks due to increased interconnectivity, demanding robust detection mechanisms. Common unsupervised methods struggle with instability from fixed threshold settings, and simple fusion of heterogeneous sensor data often yields poor results due to signal inconsistencies. This paper proposes an integration-oriented reinforcement learning (RL)-based framework combining heterogeneous data fusion, domain adaptation, and pseudo-label-guided training. Specifically, domain adaptation maps diverse industrial sensor signals into a unified latent space, extracting key features that mitigate signal inconsistencies and enable effective data fusion. These features, combined with pseudo-labels from an unsupervised anomaly detection module, constitute the state space for an RL agent. Leveraging this integrated state, the agent learns an adaptive policy-based decision rule over fused representations. To tackle reward sparsity, an Intrinsic Curiosity Module is incorporated into the RL framework, providing intrinsic rewards that encourage exploration of rare anomalies. This design addresses two major challenges: mitigating inconsistencies in heterogeneous signals (i.e. from different types of sensors) and reducing reliance on hand-tuned fixed thresholds. Experimental results show that the proposed approach improves PR-AUC by a margin ranging from 3 to 27 percentage points compared with representative baselines under the same evaluation protocol, indicating improved detection performance on the evaluated robotic anomaly-detection benchmarks.

Accurate prediction of unknown labels from feature-label datasets using machine learning is critical for applications spanning drug discovery, disease diagnostics, and climate science. However, challenges persist with limited data, high-dimensional inputs, and multi-fidelity scenarios. We developed multi-fidelity tabular prior-data fitted network (MFTabPFN), a general-purpose multi-fidelity model integrating low- and high-fidelity data through a hierarchical transformer architecture to enhance prediction accuracy and uncertainty quantification (UQ). MFTabPFN captures cross-fidelity correlations while seamlessly adapting to single-fidelity data. An active learning framework further enhances scalability by prioritizing high-value data for model refinement, minimizing resource demands in resource-intensive tasks. Evaluated on various tasks such as forest fire burned area prediction, wine quality assessment, and computational fluid dynamics, MFTabPFN outperforms state-of-the-art methods, achieving varying degrees of prediction accuracy improvement. Its versatility and robust prediction and UQ capabilities across single- and multi-fidelity datasets position MFTabPFN as a promising tool for data-driven discovery in diverse applications.

Transverse matrix cracking is the predominant failure mode in the early stages of progressive degradation, making its accurate identification crucial for ensuring the safety of carbon fiber reinforced plastics (CFRP) composite structures. However, when models trained on guided wave sensor data from one structure are applied to predict crack density in another structure with a different ply orientation, conventional deep learning methods face significant challenges due to the sensitivity of this regression task to feature scale and ply orientation. To address these challenges, we propose a domain adaptive relational graph convolutional network (DA-RGCN) model specifically designed for crack density prediction, leveraging deep domain adaptation to transfer identification knowledge learned from one laminate to another with completely different ply orientation. First, crack damage-related features are automatically extracted from sensor signals by capturing their temporal relationships during guided wave propagation. These features, along with geometric information from sensor networks, are embedded as node features within a graph structure, allowing for the learning of enhanced feature representations through the fusion of information from neighboring nodes. Subsequently, the fused features are utilized to identify crack density along each path by measuring its spatial distance from two reference states (baseline and saturation). Additionally, we employ a representation subspace distance based on principal angles to minimize distribution discrepancies between features without altering their scales. As a result, combined with the physical guidance from the damage index model, the extracted features achieve domain invariance, significantly enhancing the cross-structural generalization of the DA-RGCN. To validate the model’s capability for cross-structural identification of transverse matrix cracks, we designed transfer tasks between two layups using the CFRP Composites Dataset. The results indicate that the proposed DA-RGCN achieves an average root mean square error of 1.1841 in crack density identification, demonstrating the lowest error compared to other deep transfer learning-based and purely physics-based methods.

In many data-driven online decision systems, actions determine not only operational costs but also the data availability for future learning -- a phenomenon termed Prediction-Induced Censoring (PIC). This challenge is particularly acute in large-scale resource allocation for generative AI (GenAI) serving: insufficient capacity triggers shortages but hides the true demand, leaving the system with only a "greater-than" constraint. Standard decision-making approaches that rely on uncensored data suffer from selection bias, often locking the system into a self-reinforcing low-provisioning trap. To break this loop, this paper proposes an adaptive approach named PIC-Reinforcement Learning (PIC-RL), a closed-loop framework that transforms censoring from a data quality problem into a decision signal. PIC-RL integrates (1) Uncertainty-Aware Demand Prediction to manage the information-cost trade-off, (2) Pessimistic Surrogate Inference to construct decision-aligned conservative feedback from shortage events, and (3) Dual-Timescale Adaptation to stabilize online learning against distribution drift. The analysis provides theoretical guarantees that the feedback design corrects the selection bias inherent in naive learning. Experiments on production Alibaba GenAI traces demonstrate that PIC-RL consistently outperforms state-of-the-art baselines, reducing service degradation by up to 50% while maintaining cost efficiency.