Instrumentation & Measurement Magazine 25-2 - 13

More Efficient
and More Precise
Theoretical Models
Currently, most theoretical
models are built under
ideal conditions. For example,
TSR adopts the
standard one-dimensional
heat conduction model;
[24] put forward two standard
defect shapes (notch
and slot) by finite element
method. Meanwhile, there
are also a lot of approximations
in theoretical models. For example, during the stage
of heating phase in ECPT, the heat conduction is selectively
ignored because of the short time. In fact, the temperature
variation of natural defects is affected by many factors, which
are far more complex than the description of existing models.
Therefore, a more efficient and precise theoretical model is expected
in future research. Some work is already under way
in this regard, for example, [25] makes a detailed explanation
and an experimental analysis of the contributions of both
Joule's heating and diffusion at different times on the retrieved
thermal responses. The authors of [26] deliver a quasi-one-dimensional
heat transfer model which takes into consideration
convection heat transfer losses along the sample. The edge
crack depth of 2.5 to 6.5 mm on aluminum alloy was successfully
detected and estimated with an accuracy of 13%.
Fig. 9. (a) Test specimen; (b) Raw ECPT image; (c) Reconstructed EI image.
Towards More Accurate Quantitative Analysis
ECPT, as a qualitative identification NDT technology (i.e.,
judge whether there is defect on specimen), is quite mature.
However, on the quantitative analysis, most papers only focus
on one particular attribute of the defect, such as position,
length, width, etc. In the latest studies, some people try to
reconstruct current distribution [3], even electrical impedance
(EI) distribution [28] based on ECPT sequences images
(Fig. 9). Electrical impedance images can completely reconstruct
the profile of defects, which is beneficial to accurate
quantitative analysis. This is also a hotspot for research in
the future.
Conclusion
Machine Learning Methods Combined with
ECPT's Unique Characteristics
With the advent of the era of artificial intelligence, machine
learning algorithms provide new ideas for ECPT data analysis.
ECPT involves Maxwell's equation, heat conduction and various
complex boundary conditions under different samples,
and thus, fully accurate modeling of ECPT signal is an impossible
task at present. Machine learning algorithms are well
suited to deal with problems which are hard to solve precisely
on math. Most of the current machine learning algorithms are
based on data-driven black-box models and designed for general
scenes, which seldom consider the unique physical or data
characteristics of ECPT sequences in optimization. Therefore,
in both interpretability and time cost, there is a lot of room for
improvement. Recently, there are also relevant published papers.
For example, [27] put forward a penalty term (designed
by the characteristics of temperature response curve) in the
loss function of the traditional GAN neural network, which
improves the feature extraction ability and alleviates noises.
On the other hand, machine learning methods require a lot of
training set of data. However, it is difficult to obtain big data
on defects in ECPT. Combining the unique characteristics of
ECPT with the present popular meta-learning algorithm, generating
a large number of simulation data of ECPT defects and
even further learning a more perfect model are also potential
research directions in the future.
April 2022
In this paper, we introduce the current development of ECPT
in detail from the theoretical basis, hardware configuration
and data processing methods. ECPT is a nondestructive
testing technology combining eddy current testing and thermography,
which utilizes the thermal effect of current to detect
defects. In this sense, ECPT is an indirect measurement method.
ECPT has simple equipment configuration, high resolution,
flexible adjusted field of view, and a lot of potential for use on
rapid large-area detection. But it also needs to be pointed out
that ECPT also has drawbacks. Due to the limitation of the detection
mechanism, ECPT can only detect conductive metal
materials or some non-metallic materials (such as carbon fiber
composite materials) that can induce eddy current. Also,
ECPT has a shallow detection depth (surface or sub-surface)
due to the skin effect of eddy currents. In addition, ECPT also
faces difficulties in model construction, quantitative analysis
and high time cost of post-processing algorithms. Despite this,
ECPT is still a valuable nondestructive testing technology with
unique advantages. Recently, many researchers are committed
to further optimize and improve ECPT. We believe that as a potential
NDT technology, ECPT will play a role in more different
fields in the future.
Acknowledgment
This work was supported in part by the National Natural
Science Foundation of China under Grant U2030205, Grant
U1830207, Grant 62003075, Grant 61903065, 62003074 and
Grant U1830133.
IEEE Instrumentation & Measurement Magazine
13
Instrumentation & Measurement Magazine 25-2

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 25-2
