Unexpected fluid influx during drilling may cause serious consequence if no proper action is taken timely. A large portion of the overflow occurs during tripping. Recognizing kick during tripping automatically is difficult. This work proposes a kick detection approach for tripping operations based on unsupervised machine learning technology. By computing drill states based on the mud logging data in real-time, the kick detection algorithm is triggered for the tripping in and tripping out operations automatically. By extracting the change pattern of the tripping tank volumes during tripping operations of the historical wells, and combining the real-time data change characteristics of the analyzing well, kicks can be automatically detected. Case study shows that the proposed method can detect kick during tripping acutely and timely.