The monitoring data of smart substation devices is stored in a decentralized way. It is difficult to extract equipment defect features, so it is necessary to determine the association rules between device abnormal signals and defects by data distributed mining. Using a Hadoop framework and the MapReduce algorithm, a distributed parallel FP-growth algorithm can quickly and effectively find the strong correlation between signals. Given the defect characteristics of the secondary devices, an abnormal model is established. A method of preparing and cleaning the historical data of remote signals is proposed. This can filter out noise signals such as reset and jitter, and convert the string data into the key words data item set. The distributed parallel FP-growth algorithm is used to mine the frequent item sets and find strong correlation of abnormal signals in the historical database of each substation. The application results show that this method can effectively determine frequent abnormalities in the secondary devices and find the defects, providing the data basis for the identification of family defects. This work is supported by the National Key Research and Development Program of China “Key Technologies Research and Standardization of IoT Terminal Evaluation Platform” (No. 2018YFB21002).