🇬🇧 English | 🇮🇩 Bahasa Indonesia

SLR: Remaining Useful Life (RUL) Prediction of Reverse Osmosis Membranes with Survival Analysis & Deep Learning

A Systematic Literature Review (SLR) of 101 assessed records (79 included) on RUL prediction for reverse osmosis (RO)/desalination membranes, with emphasis on censored data (right-censored) and small data scales, as the basis for a thesis research in Information Systems.

Author: Muhammad Nio Hastungkoro (Student ID 30000226410017) - Master of Information Systems, Graduate School, Universitas Diponegoro. Contact: mnioh@students.undip.ac.id

Repository Structure

Folder	Contents
docs/	Main deliverables: SLR protocol, corpus (Excel), report, journal article, presentation, decision log, diagrams (figures/).
papers_analysis/	Per-paper analysis results (extraction.json + analysis.txt) for the included corpus.
pipeline/	Pipeline source code (search, journal classification, document building, reference verification).
metadata/	Intermediate data/JSON: master corpus, ISSN & journal quartile mapping, search statistics.
pipeline/6_fulltext_extraction_kaggle/	Kaggle kernel code (full-text extraction with Marker on GPU).
mirror_data/	SJR/Scopus data (SCImago) used for quartile classification (source: README.md).
AGENTS.md	Conventions for AI coding agents. AI-use disclosure: docs/AI_USAGE.md.

Method (summary)

Kitchenham & Charters (procedure) + PRISMA 2020 (reporting) + Webster & Watson (concept matrix)

• Wohlin (snowballing). Hybrid search; quartile classification via SCImago/SJR (Scopus derivative). Full-text extraction uses Marker (on Kaggle T4 GPU); per-paper analysis uses the Q1-Q8 quality rubric, strict-requirement dimensions (D1-D8), and relevance levels (Core-Method / Core-Domain / Core-Method+Domain / Supporting / Context).

Key Findings

• Corpus: 101 assessed, 79 included (Q1/Q2/Q3). Censoring (D3) is discussed by 28 included articles; RO domain (D7) by 2, with 0 overlap.
• Method x domain intersection (Core-Method+Domain) is only 2 articles -> a real research gap.

AI Usage

Generative AI (LLM) assisted specific documented stages (full-text extraction, thematic synthesis, drafting, docs). All study-selection and final content decisions were made by the author. Prompt templates: pipeline/prompts/; disclosure: docs/AI_USAGE.md.

Requirements

• Python 3.12+ (packages pinned in requirements.txt; version in .python-version).
• Install Python packages: pip install -r requirements.txt.
• System dependencies (not pip): LibreOffice (soffice, DOCX->PDF), Pandoc, Ghostscript (gs, for Camelot).
• Kaggle CLI + API token for the full-text extraction stage (token is not stored in this repo).

How to Reproduce

1. pipeline/1_search_and_select/: harvest metadata (OpenAlex/Crossref) -> metadata/.
2. pipeline/3_journal_classification/: quartile classification (SCImago) + mirror_data/.
3. pipeline/2_get_fulltext/ + pipeline/6_fulltext_extraction_kaggle/: obtain & extract full-text (Marker via Kaggle GPU).
4. pipeline/4_build_deliverables/: build docs/ (Excel, Word, PDF, presentation).
5. docs/figures/src/build_figures.sh: rebuild the diagrams (Mermaid + matplotlib).
6. pipeline/4_build_deliverables/build_presentation.py: validate the deck and build its PDF. (See the README in each folder. Scripts are repo-relative; adjust output paths if needed.)

Copyright Notes (IMPORTANT)

• Open Access + CC full text for 61 of 79 included (77 of 101 assessed) -> full_text.pdf included because the license permits redistribution (with attribution; see SOURCE.txt in each folder).
• The rest (24 articles: closed / bronze / unclear license) -> analysis results only (extraction.json, analysis.txt) + DOI, since there is no redistribution permission.
• OA/license status is checked automatically: pipeline/3_journal_classification/check_open_access.py (Unpaywall API) -> results in metadata/open_access_status.json.
• The code that obtains & extracts full-text is included (pipeline/, including pipeline/6_fulltext_extraction_kaggle/) so it is reproducible.

Data Attribution

Journal quartiles/classification are sourced from SCImago (SJR) - https://www.scimagojr.com (Scopus derivative). Data mirror: see mirror_data/README.md.

License

• Code (author-made): MIT-0 - free to use, no attribution (see LICENSE).
• Author-made content (documents, analysis results, metadata): CC0 1.0 - public domain, no attribution (see LICENSE-CONTENT).
• Third-party data/articles: their own licenses remain (SCImago = CC BY-NC; OA articles per their license) - see NOTICE.
• Summary: My code & content: MIT-0/CC0 (no attribution). Third-party data (SCImago, OA articles) retain their licenses.

This is a research artifact. Issues/PRs may not be monitored.