CoChat

A Robust Conformal Framework for IoT-Based Predictive Maintenance

Open access
Moccardi, Alberto, Conte, Claudia, Chandra Ghosh, Rajib, Moscato, Francesco · 2025 · Future Internet

The DOI matches its Crossref record; 32 of the 32 references it cites were found in a registry or an index; 5 of the 22 sources its claims cite were checked for support.

Abstract

This study, set within the vast and varied research field of industrial Internet of Things (IoT) systems, proposes a methodology to address uncertainty quantification (UQ) issues in predictive maintenance (PdM) practices. At its core, this paper leverages the commercial modular aero-propulsion system simulation (CMAPSS) dataset to evaluate different artificial intelligence (AI) prognostic algorithms for remaining useful life (RUL) forecasting while supporting the estimation of a robust confidence interval (CI). The methodology primarily involves the comparison of statistical learning (SL), machine learning (ML), and deep learning (DL) techniques for each different scenario of the CMAPSS, evaluating the performances through a tailored metric, the S-score metric, and then benchmarking diverse conformal-based uncertainty estimation techniques, remarkably naive, weighted, and bootstrapping, offering a more suitable and reliable alternative to classical RUL prediction. The results obtained highlight the peculiarities and benefits of the conformal approach, despite probabilistic models favoring the adoption of complex models in cases where the operating conditions of the machine are multiple, and suggest the use of weighted conformal practices in non-exchangeability conditions while recommending bootstrapping alternatives for contexts with a more substantial presence of noise in the data.

Overview

CoChat Summary

TL;DR

This paper proposes a "Robust Conformal Prognostic" framework that couples deep-learning RUL predictors with clamp-regulated (monotonicity-enforced) conformal prediction to produce distribution-free lower-bound confidence intervals for IoT-based predictive maintenance. Evaluated on the CMAPSS turbofan dataset, it benchmarks naive, weighted, and bootstrap conformal margins against probabilistic (Prophet) uncertainty quantification.

Key findings

  • On FD2, bootstrap conformal achieved the highest coverage (0.513, margin 41.66), followed by weighted (0.510, margin 40.57) and naive (0.476, margin 32.92); the authors conclude bootstrap best balances interval tightness and coverage in noisy settings.
  • Across the four CMAPSS subsets, the best point-prediction models were N-BEATS on FD1 (MSE 347.67, MAE 14.34, RMSE 18.64, R² 0.86, S-score 24069) and FD3 (MSE 226.29, MAE 10.61, RMSE 15.04, R² 0.911, S-score 22870); CNN-GRU on FD2 (MSE 300.32, MAE 13.06, RMSE 17.73, R² 0.82, S-score 75307); and BiLSTM on FD4 (MSE 505.24, MAE 17.28, RMSE 22.48, R² 0.800, S-score 273456).
  • Conformal coverage on FD1/FD3/FD4 was 0.505 (weighted, margin 40.171), 0.423 (weighted, margin 36.06), and 0.488 (weighted, margin 64.52) respectively.
  • Kaplan–Meier survival analysis set degradation-onset thresholds at cycle 125 for the single-condition datasets (FD1–FD3) and cycle 150 for the multi-condition datasets (FD2–FD4).
  • The authors report that Prophet overfits the piecewise-linear RUL target and produces "probabilistic gaps" that fail to capture sensor dynamics, whereas regulated conformal intervals adapt to degradation patterns.

Methods

  • Design: an end-to-end pipeline that (1) generates RUL predictions, (2) enforces non-increasing monotonicity ("clamp regulation"), (3) computes residuals on a calibration set, and (4) applies naive, weighted (exponential), or bootstrap margin estimators at α = 0.05 (bootstrap B = 200), selecting the margin with highest coverage.
  • Data: the NASA CMAPSS turbofan run-to-failure dataset (FD1–FD4; per Table 1, 100/260/100/249 training units, 1 or 6 operating conditions, 1 or 2 fault modes), with a unit-wise 80/20 train–validation split and a 30-lag context window.
  • Models and analysis: LSTM, BiLSTM, N-BEATS, CNN-GRU, and XGBoost compared using MSE, MAE, RMSE, R², and the asymmetric S-score (Eq. 21); conformal performance assessed via margin, coverage (Eq. 19), and EoL-proximity (Δ_index).

Limitations

  • The paper states no dedicated limitations section; it notes future work could add domain constraints, more advanced learning strategies across equipment types, and data augmentation for data-scarce regimes.
  • Evident from the text: reported conformal coverage values (0.423–0.513) fall far below the nominal 1 − α = 0.95 target, so the claimed coverage guarantee is not empirically achieved on these validation sets.
  • Evident from the text: evaluation is confined to the simulated CMAPSS dataset with no real IoT sensor deployment, and the RUL target itself is constructed ex post from a piecewise-linear/survival model rather than observed failure labels.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/fi17060244

Cites

32 of 32 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

4 Valid · 10 Not valid · 14 Not checked · 73 Own results and interpretations

20 more claims

Create a free account to see all 101 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 55
  • It may state two separate claims as one. 35
  • The quote only partly states this claim. 7

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

5 of 5

Citations

Cites

Research works printed but not linked: 2

  1. Jennions, I. (2011). Algorithms and their impact on integrated vehicle health management. Integrated Vehicle Health Management: Perspectives on an Emerging Field, SAE International. · DOI · Paper page
  2. Guo, C., Pleiss, G., Sun, Y., and Weinberger, K.Q. (2020). On Calibration of Modern Neural Networks. arXiv.

Cited by

See all 12

Metadata

Authors
Moccardi, A., Conte, C., Chandra Ghosh, R., & Moscato, F.
Affiliations
  • University of Naples Federico II (IT)
  • Department of Electrical Engineering and Information Technology, University of Naples Federico II, 80125 Naples, Italy
  • University of Salerno (IT)
Corresponding author
Moccardi, Alberto
Venue
Future Internet
Publisher
MDPI AG
Published
2025-05-30
Volume
17
Issue
6
Pages
244
ISSN
1999-5903 (ISSN-L)
OpenAlex
W4410902221
Cited by
12
References
30
Citations by year
2026: 7 · 2025: 5
FWCI
3.33
Type
journal article
Language
en
Open access
Locations
  • Future Internet (publishedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
PNRR MUR (PE0000013-FAIR)
Topics
  • Physical Sciences › Engineering › Control and Systems Engineering › Machine Fault Diagnosis Techniques
  • Physical Sciences › Engineering › Safety, Risk, Reliability and Quality › Reliability and Maintenance Optimization
  • Physical Sciences › Engineering › Control and Systems Engineering › Fault Detection and Control Systems
Keywords
predictive maintenance, uncertainty quantification, conformal prediction, remaining useful life prediction, C-MAPSS dataset, bootstrapping, confidence intervals, prognostics, Industrial Internet of Things, deep learning, machine learning

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.