EUVIMEDEUVIMED entdecken
Crossref

Kernel matrix reduction using randomization and its implementation in different kernel techniques for real time industrial process monitoring

Amir Khurshid, M Shamsuzzoha, Ajaya Kumar Pani

Measurement Science and Technology · 12.06.2026 · 37 (24) · 246201

DOI: 10.1088/1361-6501/ae69ff ↗

Kurz gefasst

Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 15.06.2026. Verfasst von Amir Khurshid und 2 weiteren Autor:innen. Die Arbeit nennt 62 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In the Industry 4.0 era, data-driven monitoring strategies are becoming increasingly significant in process industries. Principal component analysis (PCA) and independent component analysis (ICA), along with their kernelized versions, are popular techniques for linear/non-linear processes. However, kernel techniques pose computational challenges, mainly due to the need of performing eigen decomposition of a high-dimensional kernel matrix. The motive in this work is to develop a low-rank matrix approximation technique in order to address the above challenge. A new randomization strategy is proposed for kernel matrix reduction, which has been shown to work well in different kernel-based modeling techniques. The work thus proposes modified modeling techniques named randomized kernel PCA (RKPCA), randomized kernel singular value decomposition (RKSVD) and randomized kernel ICA (RKICA) for non-linear process monitoring. The proposed techniques are applied for the detection of abrupt and incipient faults associated with the process and sensors in the biological wastewater treatment process (WWTP) and exothermic reaction in a continuous stirred tank reactor (CSTR). Simulation results show that the proposed models (i.e. RKICA, RKPCA and RKSVD) show better monitoring accuracy than traditional kernel techniques of KICA and KPCA. Among the three investigated techniques, RKICA produced the maximum accuracy due to the non-Gaussian nature of the data. The Gaussianity/non-Gaussianity nature of the WWTP and CSTR data is checked by the popular Shapiro–Wilk test. Monitoring accuracy of RKICA is further improved by integrating this technique with PCA (RKICA-PCA) in order to account for both Gaussian and non-Gaussian process data with non-linearity.

Zitieren

APA

Khurshid, A., Shamsuzzoha, M., & Pani, A. K. (2026). Kernel matrix reduction using randomization and its implementation in different kernel techniques for real time industrial process monitoring. Measurement Science and Technology, 37(24), 246201. https://doi.org/10.1088/1361-6501/ae69ff

Vancouver

Khurshid A, Shamsuzzoha M, Pani AK. Kernel matrix reduction using randomization and its implementation in different kernel techniques for real time industrial process monitoring. Measurement Science and Technology. 2026;37(24):246201. doi:10.1088/1361-6501/ae69ff

BibTeX
@article{Khurshid2026,
  author = {Khurshid, Amir and Shamsuzzoha, M and Pani, Ajaya Kumar},
  title = {Kernel matrix reduction using randomization and its implementation in different kernel techniques for real time industrial process monitoring},
  journal = {Measurement Science and Technology},
  year = {2026},
  volume = {37},
  number = {24},
  pages = {246201},
  doi = {10.1088/1361-6501/ae69ff}
}
Kostenlos starten →