A Robotic System for Warped Stitching Based Compressive Strength Prediction of Marbles

SELVER M. A.

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, cilt.16, sa.11, ss.6796-6805, 2020 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 16 Sayı: 11
  • Basım Tarihi: 2020
  • Doi Numarası: 10.1109/tii.2019.2926372
  • Dergi Adı: IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Communication Abstracts, Compendex, INSPEC, Metadex, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.6796-6805
  • Anahtar Kelimeler: Feature extraction, Surface morphology, Electronics packaging, Robots, Surface treatment, Discrete wavelet transforms, Three-dimensional displays, Automated inspection systems, compressive strength (CS), machine learning, marbles, stitching, warping, CLASSIFICATION
  • Dokuz Eylül Üniversitesi Adresli: Evet

Özet

The amount, distribution, and morphology of the impurities in a marble block determine both its aesthetic quality and compressive strength (CS). Although the former property has been studied extensively, CS prediction is rarely investigated. The existing approaches either use expensive and tedious laboratory tests or employ image processing to individual surface images, which are shown to achieve limited performance. In this paper, a new electromechanical system is designed for full automatic prediction of CS of a marble block on a conveyor belt using all visible surface images, which are acquired by a three-dimensional (3-D) printed robotic arm. The images are used to generate unique reconstructions, which can represent the 3-D structure of the marbles in two-dimensional (2-D) via developed warped stitching based visualizations. Moreover, a novel feature set is proposed for taking advantage of these reconstructions. A total of 157 cubic marble blocks are collected to test the performance of the system using both conventional (neural networks) and emerging (deep) machine learning tools. Adverse effects of small sample size are compensated with data augmentation and transfer learning. It is shown that the system achieves the state-of-the-art prediction results.