The fluorine chemical industry has become an important one with rapid development because of its large variety of products, excellent performance and wide application fields. However, the hypertoxic materials which widely exist in fluorochemical engineering processes make the safety management and reliability assessment especially important. To improve the monitoring performance of the complicated fluorochemical engineering processes, an online Hidden Markov Model based method was proposed to predict the system reliability in real time. In this paper, the variables highly related to the system reliability were used as inputs for the unsupervised clustering method to classify the process states into different categories. Then they were used as the observable states to train the Hidden Markov Model with the system reliability as the hidden state. The superiority of the proposed method in estimating system reliability for the complicated fluorochemical engineering process has been strongly proved by the application on the R-22 fluorochemical engineering process. Moreover, the application on Tennessee Eastman process also confirmed its generalization performance for other complicated black-box chemical processes.