Slag grinding systems are complicated and are usually working under demanding environments. Long-term high-load operation could compromise the production of these systems, result in various malfunctions and raise the maintenance costs. In this study, with the purpose of predicting the operation conditions of slag grinding systems, a health pattern recognition system that based on data mining is proposed. Combining several algorithms, a feature filtering method is developed for analyzing the operating conditions, and for determining the indicators that affect the operations. Using healthy operating conditions as references, cluster analysis is carried out to discover the distribution of healthy conditions, and then set up a reference base. Comparing the operating data against the reference base, auto-regressive integrated moving average (ARIMA) algorithm is used to train the predicting model for forecasting the changing trends of the indicators. A corresponding software system is developed, and it is applied to real case study for proving the effectiveness and practicability of this method.