Fault Detection And Early Warning For Complex Geological Drilling Processes

With the rapid development of China,the demand for energy increases year by year,and energy resources have become an important part of our security strategy.The exploration and development of deeply buried energy resources are essential to relieve the current situation of high dependence on external energy resources.Promoting the technological innovation of deep geological exploration is a crucial problem to be solved.Complex geological drilling often goes through multiple high temperatures,high pressure,and steep structural formations.The strong downhole disturbance,nonlinear and other problems lead to the deterioration of drilling system,resulting in lost circulation,tools broken,stuck pipe,and threatening drilling safety.In the drilling process,many operating conditions,complex formation changes,and various downhole faults make fault diagnosis and early warning more difficult.With the help of data analysis and machine learning methods,this paper focuses on fault detection under multiple operation modes,incipient fault detection,early warning of drillstring faults,and multi-classification of downhole faults.The research content and innovation are summarized as follows:(1)A fault detection method for drilling fluid system under multiple operating conditions.Considering that the operation mode frequently changes in drilling processes,a fault detection method for the drilling fluid system is proposed to adapt to complex operation modes.Firstly,the distance between the real-time data distribution and the normal data distribution is calculated as a similarity index.An operation pattern recognition method is proposed based on clustering torque signal segments.Then,a normal operating zone is constructed for the standpipe pressure signal under each area.Based on the real-time similarity index and the adaptive alarm threshold under the corresponding condition,a fault detection method for loss and kick faults is proposed based on the data distribution and time series dependencies under the multi-mode.Finally,the proposed method is demonstrated by drilling process data.(2)An incipient fault detection method for the drillstring system based on the dissimilarity of drilling data distribution.The amplitude change of the drilling signal is difficult to observe in the early stage of fault.An incipient fault detection method of the drillstring system was proposed based on the difference of drilling data distribution and formation changes,so as to detect the fault of drillstring system as early as possible.First,a multivariate generalized Gaussian distribution is used to describe the distributions of real-time and historical data under normal operating conditions.The Kullback-Leibler divergence is introduced to measure the difference between the above distributions,and a new distribution difference index is defined for incipient fault detection.Then,an adaptive alarm threshold design method for the index is proposed,which makes the alarm threshold self-update with increased drilling depth.Last,the effectiveness and practicability of the proposed method are proved by industrial case studies.(3)A multi-fault classification method based on dynamic features of drilling signals under multiple time scales.Downhole faults can be reflected by multiple variables with different trends.Since capturing different variation trends at a fixed time scale is difficult,a multi-fault classification method is proposed by integrating the dynamic time series dependencies of drilling signals at multiple time scales.Firstly,a trend feature extraction method based on multi-time scales is proposed to extract various trends of original drilling signals.Then,a fault classification method based on an extended probabilistic neural network is proposed based on the extracted multi-time scale features.Finally,the superiority of the proposed method in multi-fault classification is verified by industrial process data from a geological drilling process.(4)Early warning of the drillstring faulty condition based on multi-dimensional motion model.Aiming at the complex characteristics of the drillstring system and changing formation environment,a fault warning method is proposed for the drillstring system based on multi-model fusion and self-updating strategy.Firstly,the axial and torsional motion submodels of the drillstring system are established,and the residual signal generation algorithm is proposed based on the drillstring kinematics model.To capture the early signs of fault,Wasserstein distance was introduced to detect whether the residual signal showed drastic changes.In view of the complex and changeable geological environment,an event-triggered updating scheme is proposed for drillstring models,so as to obtain the optimal model parameters in the current formation.Finally,the validity of the proposed method is verified by industrial data.This paper proposes a fault detection and early warning framework for the complex geological drilling process.It is expected to break the monopoly of foreign geological and petroleum industries on advanced intelligent drilling technology.This is important in solving the prominent contradiction between the supply and demand of resources in China.