Predicting Furnace Tube Rupture Using Multiclass Decision Forest with Explainable Risk-Based Lead-Time Classification

Furnace tube rupture is a critical safety and reliability issue in high-temperature industrial systems, often leading to unplanned shutdowns, severe economic losses, and safety incidents. Conventional monitoring approaches based on threshold alarms and single-variable diagnostics are frequently inadequate for detecting early degradation under complex multivariable operating conditions. This study proposes a multiclass predictive maintenance framework using a decision forest algorithm to predict furnace tube rupture severity from industrial operational data. The dataset comprised 10,957 observations and 56 furnace operating parameters collected from industrial historian systems. Following preprocessing and Pearson correlation-based feature selection, 19 significant parameters were retained for model development. The proposed framework integrates data preprocessing, feature selection, hyperparameter optimization, and validation using unseen operational data from 2021. The results demonstrate that the model effectively captures rupture-risk trends and provides early warning signals more than 14 days before rupture events, with high-risk classifications exceeding 70% predictive probability. Unlike conventional binary classification methods, the proposed multiclass framework enables risk-based maintenance decision-making, including targeted inspections, load reduction, and scheduled shutdown planning. The findings, based on validation against a single industrial furnace system, highlight the effectiveness of ensemble machine learning techniques in improving predictive maintenance, operational safety, and reliability engineering for this class of industrial furnace; broader generalization across furnace configurations and sites remains to be confirmed through multi-installation validation.