Furnace integrity monitoring using principal component analysis: an industrial case study

摘要

Furnace temperature monitoring, the cornerstone of furnace integrity monitoring, has traditionally been accomplished using alarm and trip limits set on individual temperature measurements of the copper coolers and refractory, with limits typically defined based on design criteria. Due to the changes in furnace operating conditions and the sheer number of temperature measurements available on a furnace, this often proves to be very ineffective. Principal component analysis (PCA) was applied to construct two models for furnace integrity monitoring: a short-term spike detection model and a long-term trend detection model. The Hotelling's 72 statistic and the lack of model fit statistic SPE were used to monitor the furnace integrity in real time, alerting plant personnel of potential abnormal process conditions. Application of the system to provide more sensitive furnace integrity monitoring and its recent use in support of a decision to safely delay the timing of a furnace endwall rebuild are demonstrated.