Automated Validation and Evaluation of Pipeline Leak Detection System Alarms

Carpenter, Philip (Great Sky River Enterprises LLC) | Henrie, Morgan (MH Consulting, Inc.) | Nicholas, Ed (Nicholas Simulation Services) | Liddell, Paul (Alyeska Pipeline Service Co.)

OnePetro 

ABSTRACT

We describe a prototype application that uses Bayesian statistics to improve Trans Alaska Pipeline System (TAPS) controller response time and reduce control room workloads by automating the process of evaluating and attributing the causes of Pipeline Leak Detection (PLD) system alarms. Pipeline Leak Detection System for Alarm Validation and Evaluation (PLDSAVE) was developed with the support of Alyeska Pipeline Service Company (APSC), operator of TAPS, and is designed to work in conjunction with the existing TAPS leak detection system.
The basis of the TAPS PLDSAVE application is an underlying probabilistic description of TAPS. This probabilistic description encompasses instrument and PLD system modeling errors, a leak condition statistical model and a description of the physical constraint conditions applicable to TAPS configuration. The stochastic model raw data is TAPS’ instrument measurements, PLD system state information, plus modeling data and leak alarms provided by the PLD system.
A Bayesian inference process is regularly applied to this stochastic model in such a way that the system determines a leak probability or validity value, and the most likely current TAPS leak and measurement/modeling error state. From the user perspective, any PLD system alarm is accompanied by the calculated leak probability value combined with an estimate of the current measurement and modeling error magnitudes that apply to the PLD system if the system is assumed not to be in a leak condition.
The goal of PLDSAVE is that TAPS safety and integrity are therefore enhanced as minimizing cost and operational distraction by (1) allowing the controller to respond in shorter time frames and assertively to high probability leak conditions, and (2) by assisting the user in the PLD system alarm attribution process by pointing to the most likely instrumentation or modeling alarm causes and conditions when a leak is of low assessed probability.