RIASSUNTO
ABSTRACT
Drillstring washout early detection is important to prevent drillstring twist-off and well blowout. In this paper, a new diagnosis method of washout failure is proposed based on a dynamic hydraulic model during a drillstring washout and pattern recognition method. The novel hydraulics model gives a detailed variation description of standpipe pressure, which can improve the washout prediction accuracy. The proposed method applied in a field well successfully recognized the drillstring washout accident at an early stage and the predicted results show a high noise-tolerant level.
INTRODUCTION
During the drilling process, harsh formation conditions and deep drilling depth are inevitable, making drillpipe failure occur frequently (Abdollahi, 2003; Albdiry, 2016). As a common type of drillpipe failure, drillstring washout is a hole or crack on drillpipe, allowing drilling mud in the drillstring flow into the annulus directly and reduce drilling efficiency (Liu, 2016). Usually, the washout holes in the drillstring develop rapidly and eventually evolve into a twist-off. Once a twist-off occurs, necessary fishing job will significantly extend the non-productive time and increasing drilling costs (Bert, 2009; Godhavn, 2010). In order to ensure economy and efficiency, an effective method for drillstring washout early diagnosis is necessary during drilling operation.
Compared with gas kick, loss and many other downhole incidents, drillstring washout has little effect on the flow rate out of the well (Owings, 1982). Therefore, it is difficult to detect a washout by monitoring pit gain variation or the difference between pump rate and flow rate out of the well. At present, many methods have been proposed to diagnose drillstring washout during drilling process, which can be divided into two categories: parameter monitoring method and model-based prediction method. In some drilling systems, washouts would be detected by threshold monitoring technique (Skalle, 2013). By installing measuring devices, washout detecting system will rise an alarm if the parameters exceeded the pre-set threshold. The problem with this method is that high noise level in the measurement usually induces high false-alarm rate. Reasonable threshold varies dramatically depending on drilling operations and measuring facilities, making it difficult to choose sensible thresholds of monitoring parameters. Furthermore, specific detecting equipment needs to be added to the system before drilling operation. In another method, model-based prediction method, a single-phase flow model under normal conditions is used to estimate the flow state of drilling fluid in the wellbore (Willersrud, 2015). A drillstring washout incident can be identified based on the difference between the estimates and measurements. Considering the error of the predicted results and measured data, the diagnosis results of model-based prediction method is not ideal during the initial stage of washout development with the small changes of parameters. To reduce the false-alarm rate, a washout incident can be detected when the deviation is large enough, which has adverse effects on early detection. Recently, the Bayesian network is applied to monitor washout incidents in real time. After sufficient training with historical data, a Bayesian network is established to model the correlation between drilling parameters and failures (Ambrus and Ashok, 2018). However, the predicted results are influenced significantly by the representation of the measured historical data.