MFL SCANNING

I am currently working with Rowan on NDE pipeline scans for a model specimen suite of pipewall anomalies.

This project is funded by the Department of Energy.

The overall goal of the project is to be able to improve accuracy and detection of pipeline defects or anomolies. We are trying to couple various sensing techniques and acheive a result that is comparable to the most informative sensor: Ultrasound. The problem with running UT "pigs" to find anomalies is that UT inspection is far more expensive to run when compared to the MFL pigs is common use today.

Below is a pipeline inspection "pig." This pig runs through pipelines and records specific pipewall character data.

© Pipeline Integrity, The "Intelligent Pig", Pipeline Integrity's inspection vehicle

Below is a photo illustrating a pipeline "pig" traveling through a gas pipeline with the fields present.

The pig contains a !! large !! permanent magnet.  When this pig is moved through the pipe, the magnet generates a large current.  A current through a conductor creates a magnetic field.  When a defect is present in the wall, the normally symetric field will become disturbed.  Flux leakage sensors around the perimeter of the pig measure this B field and the data is analyzed and plotted for the whole scanned section.

Our MFL scanning detects 3 axes of the field of our specimen anomalies.  We are currently applying 200A through the plate to provide our field.

Below is the current Scanning setup:

The raw data is imported to Matlab (high-level programming language) and converted into a grayscale matrix, DIP'ed, and inputted into a radial basis function (universal approximator)  neural net.  Here, the RBF compares 2 sensor input types and outputs redundant and complimentary information so that, eventually, we can predict the defect depth which ultimately determines whether the commonly buried pipe section need be extracted before future pipe explosions occur.

Below is an analogous model to our 2+ sensor input combination:

This analogy shows how the brain (eyes as "sensors") (in our case the ANN - Artificial Neural Net) is able to distinguish a larger object farther away from a smaller object that is closer.  By combining two sensor inputs (MFL and UT, for example) we can provide a much more accurate anomaly profile.