Document Sample
Technical Powered By Docstoc


1. Scientific and Technical Merit                                         2

     (1) Relation to solicitation objective                               2

     (2) Evidence provided to support the technology claims               2

     (3) How the proposed work will result in improvement over existing   12

     (4) Potential of scientific or engineering breakthrough              13

     (5) Scientific, technical basis, merit and anticipated benefits      13

2. Technical Approach and Understanding                                   14

     (1) Description of planned work                                      14

     (2) Labor hours and justification                                    18

     (3) Project schedule and milestones                                  19

     (4) Travel                                                           19

     (5) Technology developer                                             20

3. Commercialization Potential                                            22

4. Technical Management Capabilities                                      22


1.      Scientific and Technical Merit (Criterion 1)

(1) Relation to the solicitation objective in the targeted Topic Area of Interest

Rowan University proposes to design, develop and test a data fusion system for the

nondestructive evaluation of currently non-piggable pipes. The system will consist of a

combination of hardware and software (algorithms) that will be used to augment an

existing pipeline monitoring system that is presently being manufactured in the United

States. The augmented system will improve the accuracy and reliability of pipeline

monitoring by providing the location, size and shape of pipe-wall anomalies (defects),

without the need for an in-line pigging process.

     This proposal addresses the following applications requested in the solicitation –

     ―B. Improved technologies for internal inspection and repair of currently non-

     piggable pipes, this includes, but is not limited to: Development of new and improved

     sensors, monitors and metering devices which are capable of inspecting currently

     non-piggable pipes; new and innovative sensors for detection and characterization of

     corrosion, cracks, wall thickness, dents, gouges,……‖

(2) Evidence provided to support the technology claims

Defect characterization is a challenging task for many reasons. One of those reasons is

that sometimes there is no single testing method available to relay all the necessary

information for defect characterization. Depending on the material properties of the test

subject, one test source may provide signals with a set of limited information, say the

length and width of the defect, but may fail to reveal the depth of the defect. For the

same test specimen, a second test source may provide excellent depth information, but

the information concerning the length and width of the defect may be inadequate. In such

a case, a method of extracting the useful information from each test source and finding a

means to combine that information for defect characterization would prove superior to

the use of any one of the two sources.

       Data fusion is a method that looks to synergistically combine multiple

nondestructive evaluation signals such that vital and useful information from each signal

can be combined to improve defect characterization. Figure 1 shows the basic process

involved in data fusion. Two independent nondestructive evaluation (NDE) signals

received from the same scene or test specimen are combined in the data fusion process.

These sources can provide NDE signals from any combination of methods, for example:

ultrasound, x-rays, microwaves, thermal imaging, eddy current, magnetic flux leakage,


           Features from
            Features from
           Source 1
            Source 1                 Data Fusion
                                     Data Fusion
                                     Process                    Fused Data
                                                                Fused Data

           Features from
            Features from
           Source 2
            Source 2

                  Figure 1. Block diagram of the data fusion process.

       The resulting fused data relays two main types of information: redundant and

complementary information. Redundant information is information within each NDE

signal that is the same. For example, given three possible defect parameters, l, w, and d,

representing length, width, and depth, respectively, perhaps source one contains the

parameters l and d while source two contains the parameters w and d. The resulting

redundant information from data fusion of these two sources would be the parameter d.

Redundant information increases the accuracy and reliability in your measurements.

Complementary information is information that is different between the NDE signals

from each source. In the example discussed, the complementary information resulting

from data fusion would be the parameters l and w where the length l would come from

source one and the width w would come from source two. Complementary information

reveals features that are unique to each source and can be used to further characterize a

defect. Figure 2 shows resulting redundant and complementary information from the

data fusion process.

          Image 1                     DATA                    Redundant

         Image 2                                              Complementary

  Figure 2. Illustrating redundant and complementary information received in the data
                                     fusion process.

       The following paragraphs illustrate some preliminary attempts for enhancing the

accuracy of defect characterization using data fusion. Known defects in a prepared

specimen are inspected with using thermal imaging and ultrasonic testing; the resulting

signals are ―fused‖ using a radial basis function (RBF) neural network.

       The specimen used was a metal plate containing four small holes of the same

area, each of a different depth – 13, 28, 45, and 61 mm (See Figure 3). Ultrasound C-

scans and thermal images scans were made of the same four holes in this same specimen.

Referring back to Figure 1, the block diagram of the data fusion process, the first step in

the process is to determine two independent NDE sources or signals that provide

information about the same defect and test specimen. For our purposes, we will refer to

source one as x1 and source two as x2. For the first source, x1, information concerning the

radius and depth of each hole was extracted from the ultrasound C-scan images of the

holes. For x2, information concerning just the depth of each hole was extracted from each

of the thermal image scan images.

 Figure 3. Metal plate test specimen used for data fusion. The defects inspected appear
                                      in the box.

       Since the C-scan images present excellent data for the area of each hole

(See Figure 4), as compared to the thermal scan images, the radius of each hole

was chosen as complementary information from the ultrasound test signals. The

radius information from x1 was determined by first converting the C-scan image

of the hole to a binary or black and white image. Next, the total number of white

pixels in the image, which represent the defect area only, were totaled and

multiplied by the estimated size of each pixel (0.0625 mm2). Given this area, the

radius can be found using the formula for the area of a circle A = r2. Depth

information is also present in the C-scan images. This information was also

extracted to be used as redundant information since the depth parameter of the

thermal scan images was also available.        The peak amplitude of each raw

ultrasound C-scan was chosen to represent the depth of each defect. From Figure

5, this method is justified since the peak amplitudes of the images are

representative of the defect area in the image and fit the linear graph shown.

Figure 4. Ultrasonic C-scan images of the four small holes in the test specimen.




      Amplitude - normalized






                                   10   15   20   25        30   35   40   45
                                                   depth (mm)

Figure 5. Amplitude vs. depth plot for the Ultrasonic C-scan images showing the

validity of using the peak amplitude of each raw C-scan to represent depth.

             Similar to the C-scan data, the peak amplitudes of the thermal scan images

were chosen to represent the depth of the holes. Here there was a problem

present. It appeared that the two deeper holes had been saturated in the testing

process and only the information from shallower hole could be used. For this

reason, the project was limited to data from three of the holes. Figure 6 shows the

thermal image scans obtained for the project.

       Figure 6. Thermal scan images of the four small holes in the test specimen.

       Now that we have determined our input sources, x1 and x2, we can continue with

the next step in the process. One possible technique for fusing data is described

mathematically by the following equation:

                           f (x1, x2) = g1 (x1 (r, pa), x2 (pa)) = h (d, A)             (1)

where the source x1 is dependent upon parameters representing the radius, r, and depth or

peak amplitude, pa, of the UT C-scans an source two x2 is dependent upon the peak

amplitude of the thermal scans. The target, or solution, h(d,A) relays the defect geometry

through the defect depth, d, and the defect area, A. Finding the function g1 that uses the

two sources for input and determines h(d,A) is not a simple mathematical solution. In

fact, g1 is completely arbitrary, possibly nonlinear, and it may be impossible to represent

it mathematically. For this reason, a radial basis function (RBF) neural network can be

introduced as a universal approximator to provide the best approximation for g1.

The function approximator for g1, for a RBF network is given as:

                              g1    j  (|| xi  cij ||)                                 (2)
                                    j 1

where  denotes the weights of the hidden layer nodes in the network and  is a ―basis,‖

or window function. In this case, the radial basis or Gaussian basis function is substituted

for , given as:

                                                                  | xicij |2
                             ij (|| xi  cij ||)  e               2 2

where cij is the basis center (mean) and  is the radius (variance) of the Gaussian kernel.

Given the nature of the RBF network, this method of universal approximation can be

used to estimate the g1 function for any NDE signal. Therefore, this data fusion

algorithm can be applied to a multitude of test signals for any application given that there

is initial training data present in the form of two distinct signals x1 and x2 to characterize

the same test specimen.

       The RBF network was trained with the data extracted from both sources from the

13 mm and 45 mm defects. The 28 mm defect information was later used as testing data

to test the network. The target training data consisted of three-dimensional defect

profiles. These profiles are developed manually and used to give a visual representation

of the defect based on the predicted defect geometry. Defect characterization was

performed with the individual source data, UT only and Thermal scans only, to compare

the results to that of the data fusion algorithm.

         13 mm                        45 mm                            28 mm

                      Training                                           Test
Figure 7. Results for ultrasound data only defect characterization. Depth Predicted –
22.58 mm, Error: 19.3 %

            13 mm                        45 mm                            28 mm

                         Training                                           Test
Figure 8. Results for thermal data only defect characterization. Depth Predicted – 21.13
mm, Error: 20.95 %

    13 mm                             45 mm                                  28 mm

                    Training                                                   Test

Figure 9. Results for data fusion using ultrasound and thermal images. Depth Predicted
– 23.9 mm, Error: 14.7 %

From the results it is evident that the data fusion algorithm has some improvement for

defect characterization over using each individual source. Two attempts were made at

data fusion: one with the use of information in the form of the depth and radius of the UT

C-scans and the depth of the Thermal scans and another with only the radius of the UT C-

scans and the depth of the Thermal scans. The best results were achieved with the later

information where only the complementary information was fed to the data fusion

network, but true data fusion incorporates redundant and complementary information for

both accuracy and added features. Thus the first attempt is a true data fusion network. In

both cases, the data fusion results prove superior to the use of a single source defect

characterization network.

   In the future, improvements could be made to this project through the use of more

training data to provide more accurate results. Other data fusion algorithms (described in

a subsequent section) could be explored. Adding multiple sources, other than just the

ultrasound and thermal information, will provide more features for redundant and

complementary information and thus strengthen the data fusion network. It may even be

interesting to implement the algorithm in a case where the different sources provide a

significant amount of complementary information to see how much the data fusion

network enhances the ability to perform defect characterization on a specimen where only

one source does not provide enough information to accurately characterize the defect. In

the field, this is a cost effective means to NDE and defect characterization where

previously there is no single NDE method is able to extract the necessary information for

defect characterization.

(3) How the proposed work will result in improvement over existing technologies

The proposed sensor data fusion strategy will be employed directly to augment an

existing nondestructive evaluation system – the Local Area Monitor (LAM),

manufactured by Physical Acoustics Corporation. Developed in conjunction with the

U.S. Federal Highway Administration (FHWA), the Local Area Monitor, employs

Acoustic Emission (AE) technology for condition monitoring. Conducive to harsh

environments, the LAM has been designed specifically for remote monitoring of

structures that require hardened instrumentation. Weighing approximately 25 pounds

(with one battery pack), the system is portable, easy to handle and extremely rugged.

Originally designed for monitoring known and suspect defects on steel bridges, the

LAM’s modularity also lends itself to many other applications including the monitoring

of fatigue cracks and other discontinuities in structures, pressure vessels and

transformers. With 2 to 8 channels of digital AE, the LAM has the ability to perform

short-term condition monitoring, long-term integrity monitoring, laboratory fatigue

testing, or ―big bang‖ incipient failure detection monitoring through the use of various

user selectable options. The LAM can operate from an internal battery, external battery,

solar power, 12-16 Volt DC power or AC Voltage. Additionally, the system can be

remotely accessed via traditional phone line, or cellular phone.

In the proposed project, the existing features of the LAM will be preserved –

recommendations will be made for augmenting the system for performing multi-sensor

data fusion for pipeline inspection applications.

(4) Potential of a scientific or engineering breakthrough

The development of novel reliable sensor data fusion strategies will significantly enhance

the capability of inspecting a wide variety of infrastructure, including newer materials

such as composites that currently continue to defy conventional inspection methods. In

particular, this project will develop information theoretic based measures that will

quantify the efficacy of data fusion strategies.

(5) Scientific, technical basis, merit and anticipated benefits of the proposed work

The development of an intelligent system capable of monitoring non-piggable pipelines

by providing rapid and accurate assessment of the condition of the pipe-wall will provide

the basis for optimizing maintenance planning, and the prevention of pipeline

infrastructure failures with their attendant health, environmental, and economic hazards.

2.       Technical Approach and Understanding (Criterion 2)

(1) Description of planned work

As part of this project, Rowan University will –

        Develop a test platform for simulating pressurized pipes subjected to

         nondestructive evaluation from following methods – acoustic emission and

         thermal imaging.

        Develop novel algorithms for effectively combining data from heterogeneous

         sensors for extracting complementary information relating to the integrity of the


        Partner with the manufacturer of the local area monitoring system for

         recommending a design prototype that will employ the novel inspection


The specific aims of the proposed research project are –

     1. To augment an existing, currently manufactured, off-the-shelf pipeline inspection

         system that employs acoustic emission NDE by providing additional sensors for

         enhanced pipe-wall interrogation capabilities.

     2. To exploit recent advances in sensor technology, specifically, thermal sensors, for

         imaging pipe-wall anomalies.

     3. To design sensor data fusion algorithms that can synergistically combine defect

         related information from heterogeneous sensors for reliably and accurately

         predicting the condition of the pipe-wall.

The following tasks are planned for a successful implementation of the project objectives:

1. Design and development of the test platform

The proposed test platform is show in Figure 1. The test platform allows for biaxial

loading of test-specimens in order to simulate axial and hoop stresses in a pressurized gas

pipeline. The specimens will consist of steel coupons from actual pipeline segments and

will have simulated cracks of varying depth. A 4-channel acoustic emission system and a

thermal imaging system will be instrumented as part of the test setup.

                    AE              Specimen
                                                               Thermal Imaging


       Simulated                                                   Signal
       Defect                                                    Conditioning
                       Ram                                        Display/
                                                                User Interface

                   Figure 1: Proposed test platform with biaxial loading.

2. Design and development of the data fusion algorithms

The objective of data fusion is to combine the capabilities of each sensor modality to

provide more accurate and complete information than that available from each sensor

modality acting independently – an example of this procedure was provided earlier. An

additional capability of a data fusion algorithm is to allow for material invariant or

operational parameter invariant defect characterization. This means a single

characterization algorithm will be able to interpret defect signatures obtained from

materials with diverse characteristics and under a variety of test conditions – such

algorithms are currently being developed at Rowan for monitoring the integrity of

wastewater concrete pipelines. As part of this project, Rowan will develop measures for

evaluating the efficacy of the data fusion algorithm in terms of the information content

extracted from the various sensors. The data fusion strategy would also help direct the

evaluation of optimal sensor placement based on the configuration of the critical

component of infrastructure, determination of optimal geometry of the sensor, and

development of a system to interrogate multiple sensors.

        We propose to evaluate an incremental learning algorithm for performing data

fusion. An ensemble based incremental learning and data fusion algorithm called

Learn++ will be used. Learn++ is a new algorithm capable of incremental learning of

additional data, estimating classification confidence and combining information from

different sources. Learn++ employs an ensemble of classifiers approach for this purpose.

        Figure 11 conceptually illustrates the underlying idea for the Learn++ algorithm.

The white curve represents a simple hypothetical decision boundary to be learned. The

classifier’s job is to identify whether a given point is inside the boundary. Decision

boundaries (hypotheses) generated by base classifiers (BCi, i=1,2,…8), are illustrated with

simple geometric figures. Hypotheses decide whether a data point is within their decision

boundary. They are hierarchically combined through weighted majority voting to form

composite hypotheses Ht, t=1,…7, which are then combined to form the final hypothesis


                                                       BC1        
                                                       BC2        
          h1       h2                                                  H2
                                   h3                                                                   Hfinal
            h7                        h4                                                          

     h8                                                BC5
                    h6       h5

                                                                                            : weighted
                                                                                               majority voting
                                                       BC8              H8

                 Figure 11. Conceptual illustration of the Learn++ algorithm.

     We note that Learn++ does not specify which base classifier should be used as Learn+,

nor does it require that base classifiers be trained with identical features. In fact, the

algorithm has been shown to work with a variety of different classifiers, including different

types of neural networks on a number of practical real-world applications. In this study,

different ensemble classifiers, each trained with signals of different modalities used as

features, can be incorporated into the identification system. Such classifiers may include

neural networks, rule-based classifiers or Bayes classifiers based on acoustic emission

signals, thermal images, etc. To work in data fusion mode, Learn++ will be modified

according to structure in Figure 12, combining the pertinent information from all identifiers.

Learn++ will then provide a more informed decision, then any single identifier can



  acoustic emissions

     signals from
   different sources   AEN                                                              Final
  Thermal imaging
       data                                                            Weighted
   Data from other           TI                                      Majority Voting
                                                          Weight Assigning
                              Feature – specific
                        expert ensembles of classifiers

                       Figure 12. Learn++ for data fusion application.

3. Design of the augmented local area monitoring system

Rowan will partner with Physical Acoustics Corporation (PAC), Princeton Junction, New

Jersey, in designing a system prototype suitable for field testing. As described earlier in

this proposal, PAC currently manufactures a Local Area Monitor (LAM) for remote

monitoring of infrastructure using acoustic emission technology. Research conducted as

part of this project will generate recommendations for augmenting the LAM with thermal

sensor technology. PAC and Rowan have a history of successful sponsored collaborative

research activity.

(2) Labor hours and justification

During the 2-year duration of the project, 1 month Summer salary support is requested

each year for Mandayam (PI) and Polikar (Co-PI) – they will design and develop the test

station and data fusion algorithms respectively. Summer salary support of 0.75 onths each

year is requested for Chen (Co-PI) – he will support the thermal imaging effort. Monthly

support for 2 graduate students and hourly support (10 hrs/week) for 2 undergraduate

students is requested for performing the project tasks.

(3) Project schedule and milestones

Project Timeline

Tasks           Months     0       4         8       12      16         20       24

Test platform


Data Fusion



(4) Travel

Travel monies are requested for the following activities:

   (a) Presentation of research work for peer review at the annual conference on the

        Review of Progress in Quantitative Nondestructive Evaluation.

   (b) Periodic visits to the industrial partner.

   (c) Periodic visits to the sponsor.

(5) Technology Developer

Physical Acoustics Corporation (PAC) designs, develops, and manufactures a broad

range of Acoustic Emission (AE) instruments and systems, Ultrasonic Imaging Systems,

and Portable Eddy Current Systems. These technologies are used to identify, characterize

and visualize flaws that could lead to structural failure.     PAC also develops and

manufactures piezoelectric sensors for a variety of industries. Over the past 21 years,

PAC has attracted exceptionally talented engineers and scientists who have produced

most of the acoustic emission patents worldwide. Their accomplishments in Research

and Development have led to many innovations in the Acoustic Emission field today.

To help technology transfer, PAC has created its Engineering Services and Inspection

(ES&I) organization capable of performing basic contract research, applications and,

most importantly, Educational Training and Certification in accordance with ASNT's

SNT-TC-1A recommended practice. Additionally, PAC performs Acoustic Emission

Testing Services (field-testing) worldwide through experienced field inspectors certified

in several national and international NDT programs (ASNT Level III, PCN, COFREND,


         PAC is an employee owned and operated Company founded in 1978 and

headquartered in Princeton Junction, (bordering the University town of Princeton) New

Jersey USA. Since its founding, PAC has grown to become the acknowledged world

leader in this field. In 1980 PAC acquired Trodyne Corporation (founded in 1969) and in

1985 it acquired Dunegan Corporation (founded in 1968) which was the oldest AE

equipment manufacturer in the world. Today PAC commands better than 85% market

share of the AE equipment sales and 30% of the field testing services worldwide.

       The Company sells its products and services outside the United States through a

network of wholly owned subsidiaries and manufacturer representatives. The Company

subsidiaries are: Physical Acoustics Ltd. (PAL), EuroPhysical Acoustics S.A. (EPA),

Nippon Physical Acoustics Ltd. (NPA), and Physical Acoustics South America (PASA)

and Physical Acoustics Argentina. With PAC's solid technical experts and scientists (we

have over 500 man-years of AE experience and expertise), the Physical Acoustics Group

has maintained a refreshing stability in an NDT market that is continuously changing.

Our customer oriented mentality and business traditions make us unique problem solvers

based on solid business ethics, customer support and superior applications oriented


       PAC has two separate production facilities, housing the equipment and personnel

to meet both commercial and military specifications. Our ISO 9001 quality certification

assures PAC’s dedication to quality in design and manufacture of NDT products. Several

military specifications regarding design, production, inspection and quality assurance are

followed by PAC when specified by commercial or governmental agencies. Many of

these strict military standards are followed in the commercial production facility, due to

their increased quality and workmanship. We also have a materials testing laboratory for

R&D work, a well-equipped machine shop on location, and affiliations with research

facilities worldwide.

3.      Commercialization Potential and Benefits of the Proposed Technology to the

        US (Criterion 3)

Making a sustained profit is dependent on a number of factors. Chief among them is

reducing the product development cycle. This minimizes non-recurring engineering

development costs, which means that these costs can be recovered and profits delivered

sooner. It is crucial that universities take an active lead in exploring—and developing—

such tools. Developing partnerships between industry and universities can be an

important component to ensure the continued commercial vitality of the nation’s

industrial base.

        This project also formalizes Rowan University’s commitment to fostering market

commercialization of the R&D innovations that occur inside our labs. Taking on this new

role is in keeping with our responsibilities that now extend beyond intellectual

innovation. Through the strategic alliances we form with local industry, Rowan can

directly contribute to job creation and provide assistance to the corporate base in selected

industries. We expect this alliance to be successful: it can serve as a model to guide

similar efforts in the future.

        Since the focus of this project is on augmenting a system currently being used in

the field; and Rowan is partnering with the manufacturer of the system for arriving at a

design prototype, the potential for technology transfer into industry is high.

4.      Technical Management Capabilities (Criterion 4)

The College of Engineering at Rowan University is uniquely positioned to support strategic

alliances between academic R&D labs and local industry. Dr. Shreekanth Mandayam (PI) is

graduate from Iowa State University’s NDE group, and has been extensively involved in the

design and development of nondestructive evaluation systems, both as part of his graduate

work and as a faculty member at Rowan University. He has established a thriving NDE

laboratory at Rowan, with sponsorship from NSF, US Army TACOM, Water Environment

Research Foundation, etc. Dr. Robi Polikar (Co-PI) is also a graduate from Iowa State

University’s NDE research group. He has considerable expertise in developing defect

characterization algorithms for NDE signals and bio-instrumentation. Dr. John C. Chen

(Co-PI) is a graduate from the mechanical engineering program at Stanford University. He

has a track record of successful funded projects in the area of thermal imaging – he will

bring this expertise to the project. The capabilities of our industrial partner, Physical

Acoustics Corporation is described elsewhere in this proposal. A brief description of the

nondestructive evaluation laboratory at Rowan follows.

        Faculty research in nondestructive evaluation at the Rowan University College of

Engineering is facilitated by the ongoing development of a nondestructive evaluation

laboratory. This laboratory currently houses high precision linear motorized tables for

scanning test specimens, equipment for conducting magnetic flux leakage, DC potential

drop, eddy current, ultrasound and microwave NDE tests, along with the associated PC-

based data acquisition modules. An Ascension Technology Flock of Birds system

provides 3-D motion capture capabilities. The lab contains HP digital storage

oscilloscopes, signal generators, microwave network analyzers, UT pulser-receiver

modules and a 220A dc power supply. Pentium PCs, Sun and SGI workstations

connected to the Engineering LAN allow for on-site digital signal processing. Resident

software includes Matlab for algorithm development and Microsoft Visual C++ for

software development. A materials testing laboratory is located in the same facility. This

houses a 60,000 lb universal testing machine, a metallograph unit and provides other

destructive testing capabilities. Also in the engineering building is the machine shop,

which allows for rapid fabrication of test specimens. Furthermore, the Rowan College of

Engineering has close ties with nearby Camden Community College, that contains a

computer integrated manufacturing unit. All of the above facilities provide the Rowan

engineering faculty ready access to state-of-the-art materials characterization capabilities.

       Since its inception in 1998, the laboratory has supported funded projects (toatally

approximately 1.5 million dollars) sponsored by state and federal governments, private

foundations and industry.


Shared By: