A Machine Learning-Based Approach to the Identification of Key Material Failure Modes of the Process Facilities

Identification of credible failure modes and associated damage mechanisms is a critical part of asset integrity management program for the process units in the oil and gas industry. The Failure Mode and Effect Analysis (FMEA) has been widely used for these purposes. At the stages of units’ design and early operation, subject matter expert’s opinions (SMEs) with qualitative or semi-quantitative approaches for the damage risk assessment are often used. However, the quality of the assessment can be heavily dependent on the experts’ knowledge and information available for the studied system. The present study aims to improve the reliability of the existing credible failure mode identification by using machine learning techniques of regression and feature engineering in the framework. The developed model assesses the contribution of individual damage mechanisms to the overall component’s risk level given the available operational, design, and inspection metrics. By means of feature engineering, the limited input information is expanded for damage factor / applicable mechanism analysis by API RP 581 and API 571, used as a reference. Meanwhile, the several regression techniques will be used for risk analysis of individual damage mechanisms, in terms of their relative probabilities of occurrence and consequences. Based on which, the damage mechanisms are prioritized. The performance of the model will be tested on the example of several process units, including sour gas treatment, thermal cracking, separators, and several simulated flowlines.