Due to the data collected in the industrial processes generally have problems such as missing values, dimensions, and data imbalances, and previous classifiers are built under data balance conditions, make the classification error rate is very high, and can not actually observe we focus on the main object, which will face the challenge of fault classification.