Yudariansyah, Hadi (2025) MODEL PREDIKSI KATEGORI PERAWATAN PRASARANA JALAN REL DI PULAU JAWA. Doctoral thesis, UNDIP.

	Text Hadi Yudariansyah_21010120520007_Judul.pdf Download (9MB)
	Text Hadi Yudariansyah_21010120520007_Bab I.pdf Download (8MB)
	Text Hadi Yudariansyah_21010120520007_Bab II.pdf Download (8MB)
	Text Hadi Yudariansyah_21010120520007_Bab III.pdf Download (8MB)
	Text Hadi Yudariansyah_21010120520007_Bab IV.pdf Restricted to Repository staff only Download (8MB)
	Text Hadi Yudariansyah_21010120520007_Bab V.pdf Restricted to Repository staff only Download (8MB)
	Text Hadi Yudariansyah_21010120520007_Bab VI.pdf Download (8MB)
	Text Hadi Yudariansyah_21010120520007_Daftar Pustaka.pdf Download (8MB)

Abstract

Maintenance is the longest phase following the completion of construction and the
commencement of operations. An initial indication that a railway track requires
maintenance can be observed through the track quality values. Railway track maintenance
can be proposed based on data categorized under the Track Quality Index (TQI).
Currently, the track recording car operating on Java Island is limited to a single unit, the
EM-120, owned by PT. Kereta Api Indonesia (Persero). Over the years, the total length
of railway tracks measured annually by the track recording car has decreased. In 2019,
only 66.90% of the railway tracks were measured, leaving 33.10% unmeasured, resulting
in the absence of TQI data for certain railway sections. To address this issue, it is
necessary to predict TQI for unmeasured railway tracks to obtain comprehensive data,
enabling the submission of maintenance proposals.
This study aims to develop a predictive model for TQI categories based on manually
measured TQI data combined with the TQI dataset generated by track recording cars
across various railway sections. The proposed model seeks to provide faster, more
straightforward, and reliable results. The analysis is based on the standard deviation of
track geometry parameters, including superelevation, levelling, lining, and track gauge.
Turnouts, bridges, crossings, straight sections, and curves are classified as predictive
variables. A machine learning technique is adopted, utilizing 80% of the dataset as
training data and the remaining 20% for testing. A total of 233,175 TQI data points from
2019 to 2022 were employed to construct and validate the model.
The results of this study indicate that the predictive model for TQI Categories 2, 3,
4, and 1 achieves an accuracy of 96.09%, influenced primarily by TQI categories
variables. The remaining 3.91% variation is attributed to variables beyond the scope of
this research. The developed TQI category prediction model demonstrates statistically
significant accuracy, confirming its practical reliability for application across Java
Island's railway network.
Keywords: Categories, Track Quality Index, Machine Learning, Prediction

Actions (login required)

View Item