DEFECT DETECTION SYSTEM USING FINITE ELEMENT OPTIMIZATION AND MESH ANALYSIS

• US 20160343106A1
• Filed: 05/20/2016
• Published: 11/24/2016
• Est. Priority Date: 05/22/2015
• Status: Active Grant

• Current Assignee: The Board of Trustees of Michigan State University (Michigan State University), United States Secretary of the Army, University of Toledo
• Original Assignee: The Board of Trustees of Michigan State University (Michigan State University), United States Secretary of the Army
• Inventors: Sivasuthan, Sivamayam, Uthayakumar, Victor Karthik, Jayakumar, Paramsothy, Thyagarajan, Ravi S., Hoole, S. Ratnajeevan H.
• Granted Patent: US 10,949,584 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 30/23 using finite element methods [FEM] or finite difference methods [FDM]

Abstract

A defect detection system uses dedicated, simultaneously operating finite element optimization and mesh generation. Using an Eddy-current based probe, the system can detect and model surface and sub-surface defects.

Claims

1. An eddy current defect characterization system comprising:

an eddy current probe adapted to produce a changing magnetic field extending outwardly from the probe to induce eddy currents in an electrically conducting specimen, wherein the probe includes a sensing coil to detect a flux change from a state where there is no defect over the conducting specimen in response to the induced eddy currents;

a mesh analysis system having one or more main processors and one or more dedicated graphics processing unit (GPU) threads, the mesh analysis system having;

one or more non-transitory computer readable memories coupled to the one or more processors and to the one or more GPU threads, wherein the one or more memories include computer-executable instructions stored therein that, when executed by the one or more main processors or the one or more GPU threads, cause the one or more main processors or the one or more GPU threads to;

in response to detecting the flux in the specimen, generate, in the one or more main processors, a plurality of finite element mesh models of a defect in the specimen;

in parallel,

(i) provide each of the plurality of finite element mesh models to a different one of the GPU threads, and

(ii) perform, in each of the GPU threads, a finite element optimization on the respective finite element mesh model, and perform either (iii) or (iv)

(iii) continuously communicate the finite element optimizations from the one or more GPU threads to the one or more main processors as an input to the finite element mesh model'"'"'s mesh generation procedure;

or

(iv) evaluate, in each of the GPU threads, an objective function for each respective finite element mesh model using a genetic algorithm to identify one or more optimized finite element mesh models, and in response to the genetic algorithm evaluation, continuously communicate the one or more optimized finite element mesh models to the one or more processors to serve as an input to the finite element mesh model generation process; and

continue to generate the plurality of finite element mesh models of the defect in the specimen until at least one of the finite element mesh models satisfies a minimum objective function value.

2. The eddy current defect characterization system of claim 1, wherein the one or more memories include computer-executable instructions stored therein that causes the one or more main processors to perform a different finite element mesh model generation process such that each generated mesh model has a different mesh shape.

3. The eddy current defect characterization system of claim 1, wherein the one or more memories include computer-executable instructions stored therein that causes the one or more GPU threads to each use a matrix computation having element-by-element Gaussian iterations for the finite element optimization.

4. The eddy current defect characterization system of claim 1, wherein the one or more memories include computer-executable instructions stored therein that causes the one or more main GPU threads to determine a parametric description of the defect by optimizing a difference between (i) measured electric field/voltage profiles from the eddy current probe and (ii) computed electric field/voltage profiles from the finite element optimizations of the GPU threads.

5. A computer-implemented method of identifying a defect in a specimen, the method comprising:

(a) receiving, from an eddy current probe, measure voltage profile data from scanning the specimen for a defect, wherein the measured voltage profile data includes flux change data when the defect is encountered in the specimen;

(b) generating a plurality of finite element mesh models from the measured voltage profile data;

(c) performing, in parallel, a finite element optimization on each of the plurality of finite element mesh models using Gauss iterations;

(d) determining, in parallel and from the finite element optimizations, computed voltage profile data and determining an objective function for each of the plurality of finite element mesh models by determining a square of the difference between the measured voltage profile data and computed profiles;

(e) optimizing, in parallel, each of the objective functions using a genetic algorithm optimization; and

(f) determining the objective function having a minimum value set of parameters as describing the defect.

6. The computer-implemented method of claim 5, wherein generating the plurality of finite element mesh models from the measured voltage profile data comprises:

performing a different finite element mesh model generation process such that each generated mesh model has a different mesh shape.

7. The computer-implemented method of claim 5, further comprising using a matrix computation having element-by-element Gaussian iterations for each finite element optimization.

8. The computer-implemented method of claim 5, further comprising determining a parametric description of the defect by optimizing a difference between (i) measured electric field/voltage profiles from the eddy current probe and (ii) computed electric field/voltage profiles from the finite element optimizations of the GPU threads.

9. A non-transitory, computer-readable storage medium having stored thereon a set of instructions, executable by one or more main processors or one or more graphics processing unit (GPU) threads, for characterizing a defect in a specimen the defect detected by an eddy current probe, the instructions comprising:

instructions to, in response to detecting a change in flux in the specimen, generate, in the one or more main processors, a plurality of finite element mesh models of a defect in the specimen;

instructions to, in parallel,

(i) provide each of the plurality of finite element mesh models to a different one of the GPU threads, and

(ii) perform, in each of the GPU threads, a finite element optimization on the respective finite element mesh model, and perform either (iii) or (iv)

(iii) continuously communicate the finite element optimizations from the one or more GPU threads to the one or more main processors as an input to the finite element mesh model'"'"'s mesh generation procedure, or

(iv) evaluate, in each of the GPU threads, an objective function for each respective finite element mesh model using a genetic algorithm to identify one or more optimized finite element mesh models, and in response to the genetic algorithm evaluation, continuously communicate the one or more optimized finite element mesh models to the one or more processors to serve as an input to the finite element mesh model generation process; and

instructions to continue to generate the plurality of finite element mesh models of the defect in the specimen until at least one of the finite element mesh models satisfies a minimum objective function value.

10. The non-transitory, computer-readable storage medium of claim 9, wherein the instructions comprise instructions that cause the one or more main processors to perform a different finite element mesh model generation process such that each generated mesh model has a different mesh shape.

11. The non-transitory, computer-readable storage medium of claim 9, wherein the instructions comprise instructions that cause the one or more GPU threads to each use a matrix computation having element-by-element Gaussian iterations for the finite element optimization.

12. The non-transitory, computer-readable storage medium of claim 9, wherein the instructions comprise instructions that cause the one or more main GPU threads to determine a parametric description of the defect by optimizing a difference between (i) measured electric field/voltage profiles from the eddy current probe and (ii) computed electric field/voltage profiles from the finite element optimizations of the GPU threads.