US EPA vehicle emission test schedules are designed for non-linear dynamic system evaluation (cold start Open_Loop, transient fuel control OL, and high speed load OL). CFR required, however, bag analysis only that averaged measure berried all dynamic information (vehicle emission dynamic control behavior, especially OL). All certification and in-use tests are well engineered dyno system simulation, it's accurate and repeatable, but lark of real world monitoring approach - a loop hole for potential "Defeat Device" detecting. Since late 2015, NVFEL (National Vehicle Fuel & Emission Lab) has developed a simple device to record NOx_l_Texh during dyno dynamic driving to observe vehicle emission control behavior (as initial emission control signatures of cold start OL, TFC, DFC… ). It is also targeted to add road driving with "Signature Device" for most vehicle tests to record vehicle emission control performance in real world, which can screen out high polluter, early failure & identify "Defeat Device" existing. Road test_1: used "Signature Device" alone - non-intrusive test to detect control behaving for emission signature recondition and NOx emission (mass ratio) estimation through an Artificial Neural Network to detect "Defeat Device" (fraud). Road test_2: applied "Signature Device" + OBD II (RPM, MAFS) in order to identify high emission (mass based) real time driving conditions. NVFEL TATD currently is building the test database, then the knowledge base, and optimizing the machine learning algorithm to recognize those emission signatures and estimate emission ratios for computer to process data and make decision automatically. The system working principle, accuracy and some application test sample analysis are presented in this paper.