To improve the personalized recommendation performance of online platforms in a distributed environment while also considering user privacy protection, this paper proposes a hyperbolic attention knowledge graph recommendation algorithm for distributed architectures, integrating contextual information and privacy protection mechanisms. First, In the design of the recommendation model, a hyperbolic attention model is employed to organically integrate semantic features with an attention mechanism, thereby fully capturing the inter-relationships between products. Next, a context-aware model is designed by combining an LSTM-based memory model with AFM-based feature crossing. Furthermore, to address the aforementioned security threats, differential privacy techniques are introduced to inject noise into user features, and collaborative model training across multiple nodes is implemented within a federated learning framework. Experimental results show that, under distributed and privacy-preserving constraints, knowledge graph recommendation algorithms can achieve better precision, recall, ACC, and AUC under the same conditions. Comparing our algorithm with other algorithms, our proposed algorithm improves the ACC by 2.9 to 11 percentage points and the AUC by 3.2 to 9.1 percentage points. Similarly, our proposed algorithm achieves RMSE and MAE of 0.960 and 0.858, respectively, which can demonstrate a significant advantage. It also performs well in terms of privacy protection (such as ε-differential privacy parameters) and communication efficiency. Furthermore, system security evaluation experiments demonstrate that when the privacy budget ε ≥ 2.0, the model effectively reduces the success rate of membership inference attacks and significantly degrades the reconstruction quality of gradient inversion attacks; it also exhibits a mitigating effect against model poisoning attacks under certain conditions, thereby validating the algorithm's security and practicality in real-world distributed recommendation scenarios. Comparative results across the various algorithms show that the proposed algorithm remains the best among all models in terms of precision and recall.
INTRODUCTION: In discrete and process manufacturing and their upstream supply networks, disruptions (natural hazards, cyber incidents, logistics blockages, geopolitical actions, facility failures, and demand shocks) reach decisionmakers as fragmented text narratives, slowing shared understanding across roles, while direct text-to-image generation leaves causal structure visually undifferentiated and design knowledge tacit. The research object is information-design knowledge (an art and design discipline) applied to manufacturing-disruption communication; it is not art manufacturing. OBJECTIVES: To represent art and information-design knowledge as an explicit, inspectable v0.1 rule table and quantify the effect of a deterministic causal-organisation prompt manipulation under matched generation conditions. METHODS: The dataset comprises 1,694 RSS-harvested manufacturing-disruption narratives, of which 758 passed deterministic lexical screening. Records were embedded with sentence-transformers/all-MiniLM-L6-v2, and a six-category query-derived taxonomy was audited with HDBSCAN and sensitivity analysis. Eighteen pre-generation design-prior records were stored in SQLite. A deterministic constructor emitted matched direct (D) and causal organisation (G) prompts whose verbatim text and hashes were recorded in a CSV manifest. Stable Diffusion XL produced 96 images for 24 scenarios under two conditions and two fixed seeds; outputs were assessed with CLIP image–text cosine probes, visual-complexity descriptors, and 192 label-blinded paired judgments from two independent raters using Wilson intervals, sign tests, and chance-corrected reliability statistics. RESULTS: Clustering showed moderate taxonomy correspondence (NMI 0.578–0.613, ARI 0.456–0.592, noise 67.3–72.4%). The grammar condition won 97.4–100% of non-tied judgments across four dimensions (all sign tests p < .001; AC1 = 0.663) without improving CLIP alignment, while edge density and luminance entropy increased significantly after Holm correction; these are changes in structural complexity, not quality verdicts. CONCLUSION: Design knowledge can operate as an auditable computational control layer whose benefit registers in expert-judged communication value rather than automatic similarity scores, offering a reusable template for knowledge-conditioned industrial AIGC. The evidence concerns resilience communication artefacts, not resilience performance or decision quality, and is bounded by one generation model, 24 scenarios, and two raters.
This paper addresses a collaborative scheduling problem in distributed processing and assembly factories with multiple locations. With the objectives of minimizing the makespan, total tardiness and transportation cost, a distributed flexible assembly job shop scheduling problem with transportation is formulated, and a solution method is developed by adapting the greedy randomized adaptive search procedure, the path relinking mechanism and variable neighborhood descent to a decomposition framework. Its problem-specific elements are a three-segment encoding that represents machine flexibility and operation sequence flexibility, a topological decoding rule that guarantees precedence feasibility without repair, a factory-consistency repair that enforces the transportation constraints inside the decoder, and a whole-job factory-reassignment neighborhood that is the only move able to act directly on transportation cost. Experiments on 42 instances derived from a real case, including a factorial design of 27 instances over order size, factory network and cost structure, show that the proposed method attains a significantly higher hypervolume than MOEA/D and a deep-Q-network-based memetic algorithm. The strategy comparison shows that collaboration mainly acts on logistics: relative to independent production, collaborative production with transportation constraints reduces transportation cost by about 12% while changing the makespan by less than 1%, and the saving doubles when inter-factory transfers are expensive but falls to 2% when delivery costs dominate. Forbidding cross-regional movement of semi-finished jobs never degrades schedule quality relative to unrestricted collaboration and improves it on 36 of the 42 instances, whereas for the largest order books independent production yields better makespan and tardiness within the same computing budget.
Monitoring the progressive degradation of linear guide systems under complex operating conditions is difficult using weak vibration signals, and existing data-driven methods generally suffer from insufficient integration of physical evolution laws and poor calibration of predictive uncertainty. This issue is particularly consequential in high-precision manufacturing, where unexpected wear-induced failures of linear guides can disrupt production continuity and undermine system resilience. To address these problems, this paper proposes an AI-driven wear prediction framework that integrates physical priors with adaptive physical constraints. First, coarse-grained physical prior features are extracted from wear profile data and concatenated with PCA-reduced vibration features to establish a dual-channel fusion mechanism. Second, a lightweight residual network enhances degradation feature extraction, while adaptive physical constraints dynamically balance prediction accuracy with physical regularity. Finally, within a pseudo-Bayesian physics-informed framework, a two-stage model selection and global variance scaling strategy is proposed to achieve well-calibrated uncertainty quantification under weak signal conditions. Experimental results on a linear guide degradation dataset show that the proposed method achieves an R² of 0.9553, MAE of 101.70 μm·mm, and RMSE of 143.26 μm·mm, with a 95% prediction interval coverage rate of 0.9460. These results demonstrate that the proposed framework provides reliable, uncertainty-aware wear-state prediction and RUL-oriented decision support for the predictive maintenance of linear guides. By enabling early degradation warning, informed maintenance planning, and rapid recovery preparation, the proposed method can contribute to reducing unplanned downtime and supporting the resilience of high-precision manufacturing systems.
INTRODUCTION: Distributed industrial edge monitoring requires long-sequence models to remain reliable under non-stationary operating conditions while preventing sensitive local data from being exposed through cross-node historical-state sharing.OBJECTIVES: This work addresses secure reuse of historical representations under feature drift, privacy leakage, and poisoned-memory propagation.METHODS: We propose HSDC-Net, which locally encodes heterogeneous process streams, releases clipped and differentially private latent anchors, estimates cross-node inconsistency and recursive source trust, aligns reliable historical states through low-rank context transport, and restricts correction to a learned nuisance-drift subspace.RESULTS: Experiments on four industrial datasets show that HSDC-Net consistently improves predictive performance and drift robustness with moderate online latency. It also suppresses malicious-anchor acceptance while maintaining a finite privacy budget with only limited utility degradation relative to the non-private variant.CONCLUSION: HSDC-Net provides a unified privacy-aware and poisoning-resilient mechanism for robust distributed industrial monitoring without centralizing raw manufacturing streams across heterogeneous monitoring and maintenance tasks.
