LAUSR

Abstract Early kick detection is important to reduce the possibility of well blowouts, which have serious safety and financial implications in deepwater drilling. In this paper, a novel pattern recognition model for kick diagnosis is proposed. In the model, trend detection is used to make good decisions based upon noisy drilling data. The integrated model comprises two parts: (i) a dynamic wellbore flow model, which extracts the kick mode via multiphase flow simulation; and (ii) improved piecewise approximation and similarity measure algorithms. The proposed model successfully diagnoses gas kick faults in real time when it is applied in a field well. It offers significant sensitivity improvements while reducing the false alarm rate caused by ambiguous data, as both kick and non-kick events are accurately extracted and rapidly identified. This model is a new attempt to combat the problem of early kick detection via pattern recognition.