ultrasonic phased arrays and interactive reflectivity ... library/events/2016/fy16 cs rd/tues... ·...

Ultrasonic Phased Arrays and Interactive Reflectivity Tomography

Purdue UniversityHani AlmansouriBenjamin FosterCharles Bouman

Oak Ridge National LaboratoryHector J. Santos-VillalobosChristi JohnsonCase CollinsDwight ClaytonRoger KisnerYarom Polsky

MotivationNuclear Power Plants

Oil and Gas

Geothermal

Hani Almansouri - halmanso@purdue.edu 3

• Containment Buildings• Reactor buildings• Fuel pools• Cooling towers

CCS

Motivation: Needs

[Gasda 2004]

Hani Almansouri - halmanso@purdue.edu 4

• Low density matrix• Mixture of cement,

aggregate, rocks, and water• High density reinforcement

Motivation: Synthetic Aperture Focusing Technique (SAFT)

Non-homogenous Thick Concrete Specimen

Plexiglas

Hani Almansouri - halmanso@purdue.edu 5

Homogenous medium

Motivation: Model-Based Iterative Reconstruction (MBIR)

Hani Almansouri - halmanso@purdue.edu 6

“Rather than making the “purest” measurement, make the most informativemeasurement.” --Charlie Bouman, Professor Purdue University

MBIR looks for the best solution that fits the data

Goal

Implement MBIR to ultrasonic system for thick non-homogeneous concrete structure.

Hani Almansouri - halmanso@purdue.edu 7

Background On MBIR

Inverse Problem

Maximum A Posteriori (MAP) Estimation

Optimization

Iterations

Hani Almansouri - halmanso@purdue.edu 8

Background On MBIR: MAP Estimation

Probabilistic model of the measurements (Forward Model)

Probabilistic model of the image (Prior Model)

For 𝑊~ 𝑁(0, Λ−1) For X ~ pair-wise Gibbs Distribution

Hani Almansouri - halmanso@purdue.edu 9

Background On MBIR (cont.)

Physical System

Difference

Updating 𝑥

MBIR Simulation 𝑦 = 𝐴 𝑥

𝑦

𝑥

𝑦𝑥

𝑒

Hani Almansouri - halmanso@purdue.edu 10

Ultrasonic MBIR: Forward Model

Source IntensityIntensity Reflectivity

Coefficient (IRC)

• Homogenous, isotropic, and linear medium

• Propagation model

• is the attenuation coefficient

• is the phase delay

• Measurements

Hani Almansouri - halmanso@purdue.edu 11

Ultrasonic MBIR: Forward Model (cont.)

Hani Almansouri - halmanso@purdue.edu 12

Ultrasonic MBIR: Prior Model

• is the set of all unordered neighboring pixel pairs 𝑠, 𝑟such that 𝑟 𝜖 𝜕𝑠.

• Q-Generalized Gaussian Markov Random Field (QGGMRF)

• It has continuous first and second derivative near Δ = 0.

• for guaranteed function convexity

• Usually, p = 2 and q is close to 1.• preserves details in low contrast regions

• preserves edges

Hani Almansouri - halmanso@purdue.edu 13

X ~ pair-wise Gibbs Distribution

Ultrasonic MBIR: Optimization

• Iterative Coordinate Descent (ICD) algorithm• Fast and stable algorithm• Better for defining high frequency

components (i.e., edges)• Usually initialize object estimate with

lower resolution reconstruction• E.g., back-propagation of measured

signals 𝐴−1𝑦

Hani Almansouri - halmanso@purdue.edu 14

Forward Model Upgrade: Direct Arrival Signal

Direct arrival

Hani Almansouri - halmanso@purdue.edu 15

Forward Model Upgrade: Direct Arrival Signal

Old solution:

Hani Almansouri - halmanso@purdue.edu 16

Forward Model Upgrade: Direct Arrival Signal

New solution:

Replace Forward Model and MAP Estimate

With

Direct arrival

Hani Almansouri - halmanso@purdue.edu 17

Forward Model Upgrade: Direct Arrival Signal

New solution:

Hani Almansouri - halmanso@purdue.edu 18

Synthetic Data Reconstruction• K-Wave simulation

• 4 simulations• 2 simulations of simple phantoms

• 2 simulations of corner cases

• 10-transducer in-line system

• Input signal: 3-cycle sine wave,52 kHz central frequency

• Medium size: 40 cm wide, 30 cm thick.

Hani Almansouri - halmanso@purdue.edu 19

Synthetic Data Reconstruction No. 1

Hani Almansouri - halmanso@purdue.edu 20

Ground truth SAFT MBIR

Artifact and noise suppression

MBIR looks for the best solution that fits the data

Synthetic Data Reconstruction No. 2

Hani Almansouri - halmanso@purdue.edu 21

Ground truth SAFT MBIRThin Rectangle

Synthetic Data Reconstruction No. 3

Hani Almansouri - halmanso@purdue.edu

Ground truth SAFT MBIR

22

Synthetic Data Reconstruction No. 4

Hani Almansouri - halmanso@purdue.edu 23

Ground truth SAFT MBIR

Empirical Data Reconstruction

24Hani Almansouri - halmanso@purdue.edu

• Cement sample with a rebar inside• 10-transducer in-line system

• Input signal: 52 kHz central frequency

• Medium size: 40 cm wide, 30 cm thick.

First Experiment:

Empirical Data Reconstruction (cont.)Ground truth SAFT MBIR

25Hani Almansouri - halmanso@purdue.edu

10 Transducers equally spaced

MBIR• Identifies noise and artifacts• Easy to interpret• Clean and clear reconstruction

Empirical Data Reconstruction

26Hani Almansouri - halmanso@purdue.edu

Second Experiment:• Three cement samples

• Control with no features

• Flat steel plate

• Steel plate with ORNL text engraved and steel ball

• Field of view: 40 cm wide, 30 cm deep.

• Acquired initial data set

• The reconstruction of these samples are in progress …

Hani Almansouri - halmanso@purdue.edu 27

Closing RemarksAchievements:• Functional MBIR algorithm for ultrasonic

synthetic and empirical data.• Upgraded forward model, able to

compensate for direct arrival signal.• Better suppression of noise and artifacts

compared with SAFT.

Challenges:• Dependencies caused by frontal pixels• Multiple reflection effects

Future work:• More empirical experiments• Non-homogeneous anisotropic non-linear forward model

Ground truth MBIR