Gardado en:
| Autor Principal: | |
|---|---|
| Formato: | Recurso digital |
| Idioma: | |
| Publicado: |
Zenodo
2026
|
| Subjects: | |
| Acceso en liña: | https://doi.org/10.5281/zenodo.19705346 |
| Tags: |
Engadir etiqueta
Sen Etiquetas, Sexa o primeiro en etiquetar este rexistro!
|
Table of Contents:
- <p>Submersible pump failures in iron ore dewatering systems generate cascading costs: mechanical repairs, production downtime, and avoidable energy waste. Existing approaches treat failure forecasting and energy loss quantification as separate problems, leaving a gap between statistical anomaly detection and actionable remaining useful life (RUL) estimation. This paper proposes the Hybrid Physics-Informed ARIMA (HPI-ARIMA) framework, which closes that gap by coupling an ARIMA statistical layer with a Weibull physical degradation model through an adaptive Kalman filter. The key architectural novelty is the use of the squared standardised ARIMA residual as a real-time anomaly score that dynamically gates Kalman measurement noise — continuously adjusting the weighting between physics prior and sensor observation in proportion to statistical evidence quality. To our knowledge, this coupling mechanism has not been previously reported for submersible pump prognostics in mining applications.</p> <p>The framework is demonstrated on 17 months of operational maintenance logs (68 failure events, January 2024 – May 2025) from submersible pumps at an iron ore processing facility in Nigeria. All results are proof-of-concept estimates; performance claims are explicitly conditioned on the pilot-scale dataset and will be re-evaluated at n ≥ 24 months. A field-validated energy waste formula quantifies 178.6 kWh of cumulative fault-induced energy loss and feeds a specific energy consumption signal into the maintenance trigger logic. Risk stratification identifies 7.5 HP pumps and Collection Tank 2 as priority intervention targets. On the held-out validation split, HPI-ARIMA achieves a false alarm rate of 6.8% versus 14.2% for standalone ARIMA (52% reduction), an α-λ RUL accuracy score of 0.71, and projected monthly savings of NGN 9,000–11,900 inclusive of avoided motor repair costs. A formal INAR(1) cross-validation, ablation study, theoretical coupling derivation, and convergence analysis collectively establish the methodological foundations for the planned multi-site deployment study.</p> <p>The framework runs on standard industrial edge hardware without cloud connectivity, directly addressing the infrastructure constraints of sub-Saharan African mining operations and providing a transferable template for maintenance digitalisation in resource-constrained environments.</p>