Drill String Failure Prediction Methodology Using Data Analytics for Real Time Well Engineering

Vibration signals in the form of real-time accelerations recorded downhole often contain strong noise, making it difficult for fault or failure diagnosis during drillstring. Vibration signals include noise from sources, such as motors, bit and drillstring interactions with the borehole, rugged boreholes, and similar interactions. Sometimes, this noise is stronger than the underlying signal, which might lead to false alarms or misrecognition. This paper discusses a novel approach to diagnose failure in real-time, which is also robust to noise. Existing methods for fault or failure diagnosis are based on threshold values of peak and average vibrational signals. This paper introduces a hybrid method of combining signal demodulation with spectral analysis to predict drillstring failure. This method deconvolutes the signal with the help of minimum entropy deconvolution (MED) and Teager-Kaiser energy operator (TKEO) to remove ambiguity as a result of noise. Then, the signal is decomposed into various intrinsic mode functions (IMFs) that have the highest correlation with the original signal and can be used for failure diagnosis. This paper also discusses how spectral analysis can be applied on selected IMFs by comparing the IMF’s impact frequency with the system’s natural frequency so its harmonic drillstring failure can be diagnosed more precisely.