Open Access

ARTICLE

Developing Fault Prognosis and Detection Modes for Main Hoisting Motor in Gantry Crane Based on t-SNE

Yu-Tang Cao¹, Shin-Hung Lin¹, Raphael Brent Lau Hundangan², Edward Basaong Ang ², Tsung-Liang Wu^1,*

1 Department of Mechatronics Engineering, National Kaohsiung University of Science and Technology, Kaohsiung, 824005, Taiwan
2 School of Mechanical, Manufacturing and Energy Engineering, Mapúa University, Manila, 1002, Philippines

* Corresponding Author: Tsung-Liang Wu. Email:

(This article belongs to the Special Issue: Selected Papers from the International Multi-Conference on Engineering and Technology Innovation 2024 (IMETI2024))

Computers, Materials & Continua 2025, 84(3), 5255-5277. https://doi.org/10.32604/cmc.2025.066877

Received 19 April 2025; Accepted 18 June 2025; Issue published 30 July 2025

Abstract

The gantry crane system is a crucial equipment for loading and unloading containers at the shore site. The existing trend of crane technology is in the transition from in-site operators to remote operators outside the cargo handling site, which will comply with all safety regulations, including conditional crane monitoring. However, remote control introduces certain drawbacks in machine maintenance, as no on-site operators can provide real-time feedback on abnormalities. Therefore, this study proposes a failure detection system that uses vibratory sensors installed on machines to monitor and provide early warnings for various failures. For faulty event identification, the Fast Fourier Transform is carried out for the raw vibratory signals, and several frequency bands are classified by using t-SNE to evaluate the significance among clusters. The adjustment of hyperparameters of the t-SNE will alter the quality of the classification of different events, and this process is conventionally operated in accordance with users’ experience. In this study, we propose a novel rating approach to automatically tune the hyperparameters of t-SNE to evaluate the separation and cluster compactness of the t-SNE results. Then, the results of clusters served as input features for training the faulty event detection model, and the detection model shows more than 95% accuracy in identifying different abnormal conditions of the main hosting motor.

---

Keywords

---