This study addresses the challenges of diagnosis and prognosis in complex dynamic systems, focusing on an industrial MIG/MAG robotic welding application. A robust framework based on Recurrent Neural Networks (RNNs), specifically LSTM and GRU architectures, was developed to analyze multivariate time-series data from real-world sensor measurements for fault detection and remaining useful life (RUL) prediction. The approach incorporates a post-hoc attention mechanism to enhance interpretability by identifying the most influential variables and time windows contributing to each prediction. A sliding-window method (50 historical steps and a 10-step prediction horizon) was employed, and the models were evaluated against various benchmarks, including decision trees, feedforward neural networks, random forests, 1D CNNs, CNN–LSTM, and transformer architectures. Experimental results demonstrate that GRU and LSTM models outperform all baselines in both diagnostic and prognostic tasks, achieving up to 94% diagnostic accuracy and inference times of less than 100 ms, making them suitable for real-time deployment. The attention mechanism consistently identified key degradation signatures in more than 85% of fault sequences, offering valuable insights for domain experts. The proposed RNN-based framework therefore combines predictive accuracy, real-time performance, and interpretability, providing a scalable and effective solution for intelligent maintenance in industrial robotic systems.
23 June 2026 in Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability
Reliability growth modeling plays an important role in the analysis of repairable systems operating under evolving and non-stationary conditions; however, classical formulations such as Crow–AMSAA and Musa–Okumoto generally rely on single-regime assumptions that may limit their ability to represent heterogeneous failure dynamics and imperfect maintenance effects across successive operational phases. This paper proposes a Multi-Rate Reliability Growth Model with Partial Reset (MR-RGM-PR), a multi-regime NHPP-based probabilistic framework integrating regime-dependent dynamics and intensity-level memory propagation mechanisms to represent residual effects induced by imperfect corrective actions. An estimation framework combining maximum likelihood estimation and Bayesian inference is developed to support parameter identification and uncertainty quantification, while the theoretical analysis demonstrates that the proposed formulation preserves the principal properties required for cumulative counting processes, including continuity, monotonicity, and non-negativity, while enabling flexible representation of regime transitions and inter-regime dependency propagation. Validation experiments conducted on both synthetic and real-world datasets suggest that the proposed framework provides improved goodness-of-fit characteristics and reduced prediction discrepancies relative to several classical reliability growth formulations under the considered experimental conditions, as reflected by lower RMSE and MAE values together with improved information criteria. The results additionally highlight the potential relevance of combining adaptive segmentation and memory propagation mechanisms for representing complex reliability dynamics involving heterogeneous operational conditions and partially effective maintenance actions. From a practical perspective, the proposed MR-RGM-PR framework may support the identification of critical operational phases and provide useful insight for maintenance planning, intervention-policy optimization, and reliability-oriented decision support in industrial environments.
16 May 2025 in Engineering Applications of Artificial Intelligence