WITHDRAWN: Strain rate effect on mechanical properties of 0.24% carbon steel using artificial neural network

Khudhair, Saleh Halbus; Taher, Maher K.; Mohammed, Mortda

doi:10.1016/j.matpr.2021.03.514

Cited by 1 publication

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this area, many different machine learning approaches have been employed, for instance: support vector machine [7,8,16], neural networks with attenuation mechanisms [17], convolutional neural networks [9,[18][19][20][21][22][23], classification priority networks [24], Siamese networks [25,26], Siamese basis function networks [27], generative adversarial networks [28,29], and k-nearest neighbors [30]. Furthermore, machine learning algorithms have also been used to recognize the phases of steel [31] and steel types [32][33][34] to model the mechanical properties of steel [35]. In general, approaches based on neural networks require significant datasets for training and validation, in most cases with additional annotation for supervised learning.…”

Section: State Of the Artmentioning

confidence: 99%

Recognizing steel elements with BRDF and k-nearest neighbors

Ciszkiewicz,

Jaglarz,

Uhl

2023

Metrology and Measurement Systems

View full text Add to dashboard Cite

The paper deals with analysis of recognition of surface quality with reflective structures. Such surfaces are common in metallic materials cut using a saw or polished. There are no easy methods to identify such elements after machining. This issue is crucial in the industry for quality control as recognition of the elements, for instance after failure, allows for a detailed study of their manufacturing process. Firstly, six cuboid steel elements were obtained from a larger beam with a circular saw. Then, the bidirectional reflection distribution function (BRDF) was obtained for each element 3 times. The BRDF profiles were used in custom recognition software based on the K-nearest neighbors algorithm. In total, 140 variants of the classifier were tested and analyzed. Additionally, each variant was solved 200 times with different splits of the dataset. The results showed a high multiclass accuracy in all considered variants of the algorithm, with multiple variants achieving 100% accuracy. This level of performance was attained with only 1 to 2 training samples per class. Its low numerical complexity, easy experimental procedure, and “one-shot” nature allow for fast recognition, which is crucial in industrial applications.

show abstract