NDT identification of a crack using ANNs with stochastic gradient descent

Journal Article · Mon May 01 04:00:00 EDT 1995 · IEEE Transactions on Magnetics

DOI:https://doi.org/10.1109/20.376431· OSTI ID:63121

Arkadan, A A; Chen, Y; Subramaniam, S ^[1]; Hoole, S R.H. ^[2]

---

1. Marquette Univ., Milwaukee, WI (United States). Dept. of Electrical and Computer Engineering
2. Harvey Mudd College, Claremont, CA (United States). Dept. of Engineering

Nondestructive testing (NDT) is used to identify the anomalies and defects in inaccessible locations. Various techniques of optimization are used in NDT. In this work, the Artificial Neural Networks (ANNs) are applied with NDT to identify a crack in a conducting medium. In general, deterministic techniques are used with the back propagation algorithm (BP) to train the neural networks. The ANNs which are trained by a deterministic method have a tendency to get trapped in local minima. In this paper a stochastic version of the gradient descent is applied to train the ANNs and it overcomes the difficulties of local minima caused by the sinusoidal fields. The stochastic version used in this approach is based on the Metropolis algorithm which is frequently used in the simulated annealing.

• OSTI ID: 63121
• Report Number(s): CONF-9407177--
• Journal Information: IEEE Transactions on Magnetics, Journal Name: IEEE Transactions on Magnetics Journal Issue: 3Pt1 Vol. 31; ISSN IEMGAQ; ISSN 0018-9464
• Country of Publication: United States
• Language: English

Related Subjects

42 ENGINEERING
99 GENERAL AND MISCELLANEOUS
ARTIFICIAL INTELLIGENCE
EDDY CURRENT TESTING
NEURAL NETWORKS
STEELS
STOCHASTIC PROCESSES
TRAINING