Scholarlink

<p>Ensuring the reliability of critical industrial assets is a major challenge in Industry 4.0, particularly in multi-site environments where centralized predictive maintenance approaches are constrained by data privacy and scalability. This paper proposes a novel federated hybrid ARIMAX–LSTM framework for collaborative fan fault prognostics in the cement industry. The proposed approach combines statistical time-series modeling (ARIMAX) with deep learning (LSTM) within a Federated Learning (FL) architecture, where only the LSTM parameters are collaboratively aggregated, ensuring that raw operational data remain local to each industrial site. Experimental validation on real-world vibration data from a cement plant demonstrates that the proposed hybrid model outperforms standalone ARIMAX, LSTM, and XGBoost models, achieving an RMSE of 0.069, MAE of 0.015, and R<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> </InlineEquation> of 0.988 in centralized evaluation. In the federated setting, the global model shows stable convergence across 10 training rounds, with the mean squared error decreasing from 0.062 to 0.029, the MAE from 0.177 to 0.100, and the R<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> </InlineEquation> increasing from 0.967 to 0.983. Furthermore, client-level results confirm consistent performance gains across heterogeneous sites, with R<InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> </InlineEquation> values reaching up to 0.987, demonstrating improved generalization for sites with limited or noisy data. These results confirm that the proposed federated hybrid ARIMAX–LSTM framework provides a privacy-preserving, scalable, and highly accurate solution for time-series-based predictive maintenance in complex industrial environments, addressing a critical gap in current industrial prognostics research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Federated hybrid ARIMAX-LSTM for collaborative fan fault prognostics: A cement plant case study

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Noureddine Allassak,
  • Salima Trichni,
  • Fouzia Omary

摘要

Ensuring the reliability of critical industrial assets is a major challenge in Industry 4.0, particularly in multi-site environments where centralized predictive maintenance approaches are constrained by data privacy and scalability. This paper proposes a novel federated hybrid ARIMAX–LSTM framework for collaborative fan fault prognostics in the cement industry. The proposed approach combines statistical time-series modeling (ARIMAX) with deep learning (LSTM) within a Federated Learning (FL) architecture, where only the LSTM parameters are collaboratively aggregated, ensuring that raw operational data remain local to each industrial site. Experimental validation on real-world vibration data from a cement plant demonstrates that the proposed hybrid model outperforms standalone ARIMAX, LSTM, and XGBoost models, achieving an RMSE of 0.069, MAE of 0.015, and R \(^{2}\) of 0.988 in centralized evaluation. In the federated setting, the global model shows stable convergence across 10 training rounds, with the mean squared error decreasing from 0.062 to 0.029, the MAE from 0.177 to 0.100, and the R \(^{2}\) increasing from 0.967 to 0.983. Furthermore, client-level results confirm consistent performance gains across heterogeneous sites, with R \(^{2}\) values reaching up to 0.987, demonstrating improved generalization for sites with limited or noisy data. These results confirm that the proposed federated hybrid ARIMAX–LSTM framework provides a privacy-preserving, scalable, and highly accurate solution for time-series-based predictive maintenance in complex industrial environments, addressing a critical gap in current industrial prognostics research.