| In order to reduce machine tool’s downtime in actual production, lower maintenance costs, machine assembly quality must be detected. Among the machine mechanical parts, the feeding system is the most prone to failure. The current system for machine tool feed assembly quality detection methods are mainly experienced judge, instrumentation testing,CNC system self-diagnosis, but these methods are not yet able to accurately locate parts of the machine tool assembly fault.To solve these problems, the article proposed based on load current CNC machine tool feed system assembly quality detection methods. The main principles of the methods involved are common machine tool feed system assembly fault, the relevance of the forces and load current fault signal under each of the three, providing a feasibility analysis based detection methods by relevance.According to the principles of experimental design, we use a single set of control experiments were carried out four experiments: standard data method, shield fault, excessive preload, screw rails’ not paralleling. After collecting load current data, we can get the fault signal obtained features by analysis. Shield fault current spikes, excessive preload tilts at both ends, screw rails not parallel to the current adjustment tilts at one side.The collected data are preprocessed and the feature vectors are normalizd to get model which is based on multi-classification DAG-SVM algorithm. 26 groups of test data are introduced for the model training, which is aim at obtaining a optimal decision surface of classification. Ultimately, the three common failures could be successfully identified, the comprehensive accurate rate is 96%.The method mentioned above is applied to HNC-8 CNC system for the secondary development of detection module which uesd for the feed system of assembly quality. Four groups of test data are introduced and it is successful to verify its accuracy. Currently, thismethod is available to accurately diagnose three common feed system failures, which are shield fault, excessive preload, screw rails’ not paralleling respectively. Operating procedures are provided for routine testing maneuver before leaving factory and detecting during usage. |