Abstract The stability of product quality is a crucial issue in the process industries. Quality-related fault isolation usually assists in the real-time monitoring of process industries, thus allowing for better product quality and higher economic benefits. However, quality variables are usually difficult to be measured online due to economic and technical constraints, which makes traditional fault isolation methods inadequate for quality-related faults. In this work, an extensible quality-related fault isolation framework is proposed based on dual broad partial least squares (DBPLS). First, broad learning system (BLS) is integrated with partial least squares (PLS) to develop the offline model. Then, just-in-time-learning (JITL) PLS based on a new weighted similarity index is used for online soft sensor modeling. After that, the extensible version of DBPLS is given when new faults occur. In addition, an alternative formulation of DBPLS are further discussed. Finally, the proposed framework is applied to a real hot rolling process, where it can be found that DBPLS can extract the valuable information from the process variables related to quality variables and have better classification performance than other existing methods.