Uncertainty-Aware Wall-Thickness Forecasting and Risk-Based Inspection Scheduling for Slurry Pipelines Using Multi-Rate Ultrasonic and Process Historian Data

Slurry pipelines in mining and mineral-processing systems are exposed to coupled erosion–corrosion mechanisms that progressively reduce wall thickness and can lead to integrity failures if inspections are not timed appropriately. While machine-learning models can predict thickness loss from operational data, point predictions alone are insufficient for risk-based inspection (RBI), where decisions depend on the probability of violating a minimum allowable thickness under measurement noise, process variability, and model error. This paper proposes an end-to-end uncertainty-aware framework that integrates multi-rate data alignment, probabilistic thickness forecasting, and decision-oriented RBI scheduling. High-frequency process historian signals (e.g., flow, pressure, temperature, pH, and solids proxies) are aggregated over lookback windows and combined with degradation-state features to construct aligned covariates for low-frequency ultrasonic thickness measurements collected at multiple sensor locations. Uncertainty is quantified through three complementary strategies: conditional quantile regression for heteroscedastic prediction bands, distributional gradient boosting that learns feature-dependent predictive distributions, and conformal calibration that provides model-agnostic prediction intervals with empirical coverage guarantees under rolling deployment. The calibrated predictive outputs are translated into actionable integrity metrics, including probability-of-limit-crossing and time-to-threshold indicators, and are used to determine inspection timing via an explicit cost–risk objective. A comprehensive experimental protocol is presented with time-respecting evaluation, calibration diagnostics, proper scoring rules, and RBI utility measures, demonstrating that uncertainty-aware methods improve the reliability and auditability of inspection decisions even when point accuracy gains are modest. The proposed framework provides a practical blueprint for deploying trustworthy, data-driven RBI in industrial slurry pipeline monitoring systems.