Noise is prevalent in all pipeline instrumentation and measurement system data and has been shown to affect Computational Pipeline Modeling (CPM) based leak detection (LD) systems in practice. Noise present in data can be random or repetitive, occurring continuous or in isolated bursts and can have a multitude of sources. At Enbridge, the leak detectability of CPM LD system is notably affected by the presence of noise in the pipeline data. Data smoothing of noisy pipeline data is a non-trivial process; these techniques are used not only to eliminate "noise" from the operating data, but also to extract real trends and patterns within the data, and to maintain important characteristics of the signal itself, particularly during transient time periods.
The data smoothing initiative began within the Leak Detection group around August 2011 and the wavelet transformation technique was adopted as the data smoothing methodology to remove or attenuate the noise present in the pipeline data to improve the leak detectability of current LD system (MBS or Mass Balance System).
Extensive offline studies and pilot tests including simulated leak tests, fluid withdrawal tests, and API 1149/API 1130 calculations have been carried out to prove this concept . The offline studies have shown promising results and Enbridge has decided to pilot this technology with real time SCADA data in 2013. If successful this pilot will be the catalyst for the full implementation of this technology on all Pipelines.
Online implementation is a non-trivial process as it includes not only modifications to the real time communication architecture between the SCADA system and the leak detection CPM, but also includes code integration, windows service development, and dealing with performance issues arising from the online assessment and evaluation.
To properly assess performance it has also been commited to performing a full fluid withdrawal test during this Pilot to assess real time performance of Data Smoothing.
This paper reviews the data smoothing methodology employed, and discusses the issues faced during the online implementation as well as the online test results. Extensive pre-implementation and online tests have been carried out to evaluate and assess this methodology. These test results indicate that a data smoothing technique has a substantial effect on the data quality which in turn has improved the detectability of current leak detection system.