Dent Safe Excavation Pressure and Fracture Reliability Assessment With Machine Learning

Abstract Dents can potentially cause cracking or pipeline leaks posing a threat to pipeline integrity. Engineering critical assessment of dents interacting with cracks plays an important role in the risk and integrity management of pipelines. It is not an easy task to conduct fracture assessment on a dent-crack defect due to the complexity of the stress field in the dent area. The failure assessment diagram (FAD) method has been widely used to conduct elastic-plastic fracture mechanics analysis of structural components [1, 2]. The FAD approach can be used to conduct fracture assessment for an arbitrary through-wall thickness stress distribution in a pipe with cracks. Finite element simulations are normally required to extract the stress distributions for a pipeline dent. However, the process is computationally demanding and tedious. A machine learning-based methodology has been developed to improve the efficiency of the process of dent fracture assessment. The machine learning algorithms were trained using a dataset built from finite element simulations of dents. Through-wall thickness stress distributions at multiple critical locations with potential cracks in the dents were extracted for varied dent geometries at multiple pressure levels. The extracted stress distributions were used to calculate stress intensity factors (SIFs) for a wide range of crack dimensions based on the SIF solutions obtained from API 579 [3]. The output variable of the machine learning model was defined as the ratios of SIFs with and without dents. The inputs to the machine learning algorithms to train on include pipe and dent geometries, crack dimensions and pipe internal pressures. The model performance was evaluated through a three-way holdout method. It was shown that the trained machine learning models were able to effectively predict SIFs for dent-crack defects with varied dent geometries. The models were also shown to be robust when used to conduct Monte-Carlo simulations as part of a reliability assessment to develop a distribution of SIF accounting for uncertainties on the model input variables. Dent fracture assessment with the trained machine learning models was employed in the determination of safe excavation pressure (SEP) based on a reliability analysis. The fracture assessment was conducted using the API 579 FAD approach. The effectiveness of the FAD method on the prediction of burst pressure limits for dent-crack defects was validated with full-scale tests. It was then shown that the machine learning model-based approach was effective in conducting reliability-based fracture assessment on dent-crack defects with the FAD method to determine SEPs corresponding to given reliability targets. The efficiency on the process of SEP determination has been significantly improved compared with the surrogate modeling approach using stress distributions extracted from the finite element simulations [4].