Sunil Tyagi, PhD

Publications

Preprint, patents & papers

Every paper on my Google Scholar profile, with the full text to download. Also indexed on ORCID and Scopus.

Preprint · 2026

SKANN: deep learning for vessel re-identification

Underwater acoustic target recognition has largely settled on classifying vessels by type, which does not answer whether a monitoring system has heard a particular hull before. This paper formalises that question as open-set, cross-passage re-identification on public hydrophone data, and specifies an evaluation protocol that closes the easiest routes to a flattering score — hull-disjoint splits, galleries and queries drawn from separate passages, and an audio-adjudicated transit-deduplication gate. Transit deduplication alone removes a 16–21 point apparent rank-1 advantage, larger than any difference between the methods compared. The results support analyst triage over a ranked shortlist, not identification.

Intellectual property

Patents

Granted · Indian Patent Office · 2016

A Machine Learning Tool for Ball-Bearing Fault Detection. Ball bearings are fundamental components in rotating machinery, yet their failure remains one of the leading causes of unplanned downtime and catastrophic machinery failures. Detecting faults at an incipient stage — before they escalate into critical issues — is essential for minimising disruptions and extending equipment life.

However, early fault detection is challenging because the vibration signatures of a healthy bearing and one with an emerging defect often appear strikingly similar. Common faults, such as cracks or pits on the outer race, inner race, or rolling elements, may initially generate subtle changes that are difficult to distinguish using conventional diagnostic methods.

Cite as: Tyagi S. Bearing Fault Detection. Official Journal of the Indian Patent Office, p. 20909, 27 May 2016. Patent Application No. 201621003344 A · IPO Grant No. 373427 · Named inventor; patentee DIAT Pune · Patent document (Figshare)

Demonstration — the patented bearing-fault detection tool.
Provisional · Indian Patent Office · filed September 2026

System and method for open-set re-identification of individual vessels from underwater ship-radiated noise using a raw-waveform selective-kernel acoustic neural network, and a cross-passage evaluation protocol therefor.

SKANN is a raw-waveform neural network for open-set re-identification of individual vessels from underwater ship-radiated noise. It builds a hull-identity embedding through multi-resolution selective-kernel processing, then matches a recording against an enrolled gallery by cosine similarity and rejects hulls that are not in it.

The application also covers the leakage-resistant cross-passage evaluation protocol and the transit-level deduplication of source recordings that supports it.

Provisional application, Indian Patent Office. Application No. 202611107132, filed 6 September 2026 · Applicant: Oravont Systems LLP · Inventor: Sunil Tyagi · Complete specification pending.

Patent graphical abstract: hydrophone waveform, selective-kernel encoder, hull-identity embedding Graphical abstract — raw waveform to hull-identity embedding, gallery match or rejection, with the cross-passage evaluation protocol that keeps the measurement leakage-free.
Edited

Books

Contributed

Book chapters

Peer-reviewed · machinery diagnostics & applied ML

Journal papers

The 2014–2017 papers arise from PhD research at the Defence Institute of Advanced Technology, Pune. Indexed records: ORCID 0000-0001-5897-7955 · Scopus Author ID 57194035292 · Google Scholar.

Earlier work · bearing diagnostics

Conference & early papers

Defence & strategic affairs

Articles & commentary

From the bench

Working notes