OBJECTIVE

Offshore pipeline operators are bound by PIMS requirements to inspect threats that contribute to the pipeline risk. Specifically, for freespan threat, operators often encounter dynamic freespan behavior due to uneven seabed, wave-current scouring actions and other causes that lead to change in freespan locations and dimensions. Inability to gain insight on the dynamic freespan behavior triggers frequent underwater inspection mobilization to improve visibility on the freespan condition.

METHODOLOGY

Suitable pipeline candidates with huge number of freespans reported throughout their operation lives, are selected as pilots to develop a prediction tool with an objective to predict dynamic freespan behavior. The freespan data along with other parameters such as metocean data, bathymetry profiles, pipeline coordinates and soil data are fed into a deep learning model.

Tapping into the big data means to retrospect into records dated back to the pipeline installation year. Most of them are not in digitize format and require data pre-processing before its value can be mined. A thorough data pre-processing comprises of four stages namely data extraction, data cleaning, data interpolation and data wrangling. A deep-learning algorithm was built to learn about what has happened in the past to predict future events starting from data normalization, data splitting, model training, testing and validation.

RESULTS

The prediction results which provide the freespan locations and dimensions are measured against performance metrics i.e. R-squared (R²) and Mean Absolute Error (MAE). The algorithm able to produce predicted freespan length with R² more than 70% based on deep learning using data from five pipeline candidates within the same region.

This provides operators the ability to predict freespan behavior and grasp the dynamicity in freespan dimensions between each inspection intervals which was previously unknown.  As a result, pipeline operators able to prolong underwater inspection and freespan rectification campaign to realize 20% OPEX optimization.