Since 2009, hundreds of millions of smart meters have been installed nationwide. State Grid Corporation of China manages the asset by centralized inspection in large scale. Meanwhile, a system called Power User Electric Energy Data Acquire System was built up to monitor the operation status of these meters. Therefore, vast amounts of life-time data of the meters can be obtained from the centralized inspection system as well as the data acquire system. This paper proposed a way of establishing the connections between fault data and inspection data by training neural network under MATLAB environment. And then the possibility and effectiveness to obtain fault information of smart meters in advance by data mining is discussed, to provide an alternative way of quality control of smart meters.