Abstract: In order to provide an early warning for equipment failures in primary processing line, a method for identifying the failure trend of equipment was proposed based on SOM neural network. The key process variables in primary processing line were categorized and five models were established separately for running status monitoring, including no obvious trend, ascending trend, descending trend, positive step trend and negative step trend. The ascending/descending trend referred to the slow but abnormal variations of equipment running status induced by the ascending/descending of process variables; and the positive/negative step trend referred to the instantaneous and noticeable variations of equipment running status caused by the sudden variations of process variables. The established models were trained by SOM neural network to realize on-line monitoring of equipment running trend. The said method was compared with Shewhart control chart, the results indicated that within 30 working days, the effective abnormal trend detected by the proposed method was 38 times more than that detected by Shewhart control chart, and the accuracy of early warning increased by 33%. The proposed method provides a technical support for the promotion of intelligent failure diagnosis.