FedCMAPSS: A Benchmark for Federated Learning in Remaining Useful Life Estimation

Amelia Sorrenti; Matteo Pennisi; Concetto Spampinato; Simone Palazzo

cs.LGAug 26, 2026

FedCMAPSS: A Benchmark for Federated Learning in Remaining Useful Life Estimation

Authors: Amelia Sorrenti, Matteo Pennisi, Concetto Spampinato, Simone Palazzo

Organizations: University of Catania, Italy

Abstract

Data-driven prognostics and health management has emerged as a key enabler for Industry 4.0, yet the development of robust remaining useful life (RUL) estimation models is often limited by the scarcity of run-to-failure data. While federated learning offers a promising paradigm to collaboratively train predictive models without sharing sensor data, research efforts have operated so far in the absence of a common evaluation framework. To address this gap, this paper introduces FedCMAPSS, a benchmark for federated RUL estimation based on the commonly-used NASA C-MAPSS dataset. We define a set of five standardized tasks designed to simulate real-world industrial challenges, ranging from ideal IID settings to extreme statistical heterogeneity, and conduct a systematic evaluation of state-of-the-art federated optimization algorithms across multiple neural architectures. By establishing reproducible baselines and making the source code and data splits publicly available, this work aims to provide a standard foundation for developing and comparing federated predictive maintenance solutions.

Figures & tables

	Task A	Task B	Task C	Task D	Task E
LSTM	21.53 $\pm$ 0.57	21.27 $\pm$ 0.14	28.66 $\pm$ 0.20	22.58 $\pm$ 0.09	37.16 $\pm$ 0.59
AFT	40.40 $\pm$ 0.14	37.26 $\pm$ 0.38	46.14 $\pm$ 0.03	39.93 $\pm$ 0.03	42.08 $\pm$ 0.58
AttBiGRU	25.47 $\pm$ 0.88	22.37 $\pm$ 0.19	23.04 $\pm$ 0.36	33.05 $\pm$ 0.07	36.94 $\pm$ 0.61
RNN	40.34 $\pm$ 0.10	43.29 $\pm$ 0.04	45.27 $\pm$ 0.03	39.82 $\pm$ 0.02	41.22 $\pm$ 0.51
CNN	40.88 $\pm$ 0.21	37.77 $\pm$ 0.51	46.47 $\pm$ 0.04	39.37 $\pm$ 0.22	40.82 $\pm$ 0.47

TABLE I : Local client performance in the isolated setting (RMSE).

	FedAvg		SCAFFOLD		FedDyn		FedCross
	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$
LSTM	17.66 $\pm$ 1.11	0.99 $\pm$ 0.43	17.36 $\pm$ 1.11	0.81 $\pm$ 0.27	23.45 $\pm$ 0.72	2.38 $\pm$ 0.63	18.81 $\pm$ 1.08	1.06 $\pm$ 0.53
AFT	17.19 $\pm$ 1.09	0.81 $\pm$ 0.26	16.59 $\pm$ 1.41	0.88 $\pm$ 0.49	43.71 $\pm$ 1.12	9.78 $\pm$ 1.86	18.79 $\pm$ 1.30	1.47 $\pm$ 0.93
AttBiGRU	17.84 $\pm$ 1.01	1.14 $\pm$ 0.76	17.46 $\pm$ 1.67	0.92 $\pm$ 0.45	31.32 $\pm$ 3.17	12.79 $\pm$ 6.43	22.37 $\pm$ 1.59	2.95 $\pm$ 3.72
RNN	18.97 $\pm$ 0.81	1.10 $\pm$ 0.28	18.96 $\pm$ 0.88	1.11 $\pm$ 0.29	41.82 $\pm$ 0.31	6.89 $\pm$ 0.42	20.01 $\pm$ 0.80	1.34 $\pm$ 0.41
CNN	14.79 $\pm$ 0.58	0.39 $\pm$ 0.14	14.56 $\pm$ 0.55	0.40 $\pm$ 0.13	40.45 $\pm$ 0.41	6.62 $\pm$ 1.33	15.48 $\pm$ 0.76	0.39 $\pm$ 0.07

TABLE II : Results on Task A by architecture and federated method.

	FedAvg		SCAFFOLD		FedDyn		FedCross
	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$
LSTM	21.14 $\pm$ 0.58	9.85 $\pm$ 2.20	18.42 $\pm$ 0.38	7.06 $\pm$ 0.79	40.83 $\pm$ 1.61	$\sigma\approx 36$	16.43 $\pm$ 0.31	5.35 $\pm$ 1.25
AFT	29.30 $\pm$ 3.22	$\sigma\approx 185$	17.30 $\pm$ 0.86	6.11 $\pm$ 1.81	47.17 $\pm$ 3.19	$\sigma\approx 60$	15.85 $\pm$ 0.63	3.87 $\pm$ 0.79
AttBiGRU	27.17 $\pm$ 1.35	$\sigma\approx 40$	17.84 $\pm$ 0.37	5.96 $\pm$ 0.56	38.24 $\pm$ 2.32	$\sigma\approx 530$	16.90 $\pm$ 0.38	4.49 $\pm$ 0.79
RNN	28.03 $\pm$ 2.74	$\sigma\approx 87$	20.06 $\pm$ 0.37	11.99 $\pm$ 1.92	44.82 $\pm$ 0.35	67.59 $\pm$ 4.09	20.27 $\pm$ 0.51	10.45 $\pm$ 2.88
CNN	46.78 $\pm$ 3.02	$\sigma\approx 370$	37.30 $\pm$ 1.71	$\sigma\approx 86$	42.26 $\pm$ 0.34	49.42 $\pm$ 6.79	33.69 $\pm$ 1.31	$\sigma\approx 206$

TABLE III : Results on Task B by architecture and federated method.

	FedAvg		SCAFFOLD		FedDyn		FedCross
	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$
LSTM	29.48 $\pm$ 0.34	38.26 $\pm$ 4.96	27.31 $\pm$ 1.05	31.73 $\pm$ 21.26	39.92 $\pm$ 2.40	15.01 $\pm$ 4.61	17.28 $\pm$ 0.92	1.44 $\pm$ 0.96
AFT	30.32 $\pm$ 0.33	41.59 $\pm$ 5.97	31.22 $\pm$ 1.91	$\sigma\approx 55$	49.44 $\pm$ 1.26	48.12 $\pm$ 8.59	17.75 $\pm$ 0.70	1.35 $\pm$ 0.62
AttBiGRU	30.87 $\pm$ 0.24	27.14 $\pm$ 7.59	32.15 $\pm$ 2.39	$\sigma\approx 50$	36.30 $\pm$ 2.11	$\sigma\approx 95$	20.18 $\pm$ 0.68	1.89 $\pm$ 0.60
RNN	30.52 $\pm$ 0.50	27.10 $\pm$ 5.97	34.69 $\pm$ 4.51	$\sigma\approx 41$	45.24 $\pm$ 0.28	23.75 $\pm$ 1.27	18.24 $\pm$ 0.33	1.32 $\pm$ 0.34
CNN	35.12 $\pm$ 0.38	$\sigma\approx 39$	33.10 $\pm$ 0.59	$\sigma\approx 58$	42.93 $\pm$ 0.20	17.76 $\pm$ 2.79	26.22 $\pm$ 0.68	$\sigma\approx 55$

TABLE IV : Results on Task C by architecture and federated method.

	FedAvg		SCAFFOLD		FedDyn		FedCross
	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$
LSTM	14.12 $\pm$ 0.65	4.23 $\pm$ 0.61	12.58 $\pm$ 0.25	3.85 $\pm$ 0.21	24.67 $\pm$ 1.05	14.51 $\pm$ 4.28	12.28 $\pm$ 0.23	3.52 $\pm$ 0.31
AFT	16.30 $\pm$ 0.64	$\sigma\approx 90$	15.64 $\pm$ 0.21	$\sigma\approx 17$	39.86 $\pm$ 0.38	$\sigma\approx 20$	15.44 $\pm$ 0.41	$\sigma\approx 67$
AttBiGRU	15.15 $\pm$ 0.25	4.89 $\pm$ 0.57	14.13 $\pm$ 0.08	5.20 $\pm$ 0.50	33.84 $\pm$ 0.93	$\sigma\approx 26$	14.05 $\pm$ 0.15	4.67 $\pm$ 0.40
RNN	13.84 $\pm$ 0.38	5.38 $\pm$ 2.11	15.63 $\pm$ 0.11	4.37 $\pm$ 0.14	40.06 $\pm$ 0.33	67.82 $\pm$ 14.08	15.05 $\pm$ 0.23	4.13 $\pm$ 0.21
CNN	17.15 $\pm$ 0.31	6.55 $\pm$ 1.40	17.90 $\pm$ 0.38	3.90 $\pm$ 0.34	40.24 $\pm$ 0.71	73.70 $\pm$ 30.43	15.40 $\pm$ 0.22	3.72 $\pm$ 0.34

TABLE V : Results on Task D by architecture and federated method.

	FedAvg		SCAFFOLD		FedDyn		FedCross
	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$	RMSE	NASA $\times 10^{-3}$
LSTM	27.30 $\pm$ 0.82	5.42 $\pm$ 1.01	28.92 $\pm$ 0.99	9.63 $\pm$ 1.70	34.41 $\pm$ 1.70	32.86 $\pm$ 11.42	28.15 $\pm$ 1.10	7.83 $\pm$ 1.73
AFT	34.42 $\pm$ 1.26	15.70 $\pm$ 3.65	34.81 $\pm$ 1.30	20.16 $\pm$ 4.52	41.00 $\pm$ 0.45	6.73 $\pm$ 0.45	36.94 $\pm$ 1.41	35.02 $\pm$ 6.60
AttBiGRU	32.24 $\pm$ 0.80	8.79 $\pm$ 1.25	32.74 $\pm$ 0.86	12.47 $\pm$ 1.68	42.29 $\pm$ 3.65	$\sigma\approx 90$	33.13 $\pm$ 0.83	13.10 $\pm$ 2.80
RNN	41.11 $\pm$ 0.86	5.02 $\pm$ 0.34	40.01 $\pm$ 0.64	4.31 $\pm$ 0.16	41.10 $\pm$ 0.90	5.02 $\pm$ 0.37	40.44 $\pm$ 0.74	4.52 $\pm$ 0.26
CNN	29.69 $\pm$ 0.59	6.10 $\pm$ 1.27	30.35 $\pm$ 1.01	11.53 $\pm$ 2.92	40.25 $\pm$ 0.51	5.11 $\pm$ 0.29	32.69 $\pm$ 0.80	14.47 $\pm$ 3.84

TABLE VI : Results on Task E by architecture and federated method.

Panel of Figure 7 — Fig. 1 : Average RMSE across all architectures for different numbers of local training epochs per round.

Panel of Figure 8 — Fig. 2 : Average RMSE across all architectures over communication rounds, using 10 local epochs per round.

Panel of Figure 9 — Fig. 3 : Aggregated RUL trajectories over the life percentage.

Federated Learning

Federated learning trains shared models across data holders without requiring them to pool their raw data.

See all 938 papers

Collaborative System Failure Prognostics via Federated Longitudinal-Survival Modeling
Jul 28, 2026Fan Yang, Madelyn Weller, Dimuthu Fernando +2RUL Estimation Survival Analysis
Fowler School of Engineering, Chapman University, Orange, CA 92866 USA · Department of Statistics, Grand Valley State University, Allendale, MI 49401 USA
Integrating Survival-Based Aging Models with Data-Driven RUL Prognostics
Oct 5, 2026Abhishek Srinivasan, Juan Carlos Andresen, Sepideh Pashami +1RUL Estimation Predictive Maintenance
Traton AB, Södertälje, Sweden · Halmstad University, Halmstad, Sweden · RISE Research Institutes of Sweden AB, Stockholm, Sweden +1
Robust and Personalized Federated Learning for Aircraft-Engine Prognostics under Benign and Adversarial Client Heterogeneity
Aug 4, 2026Chinmoy Mitra, Md. Mehedi Hasan Nipu, Mohammad Sakib Mahmood +2Federated Learning Aggregation RUL Estimation
Department of Computer Science and Engineering, Rajshahi University of Engineering & Technology, Kazla, Rajshahi, 6204, Bangladesh · Department of Electrical and Computer Engineering, North South University, Dhaka 1229, Bangladesh. · Department of Computer Science, Missouri State University, Springfield, MO, 65809, USA. +2

Jul 28, 2026J/K move · Enter open · S save

Collaborative System Failure Prognostics via Federated Longitudinal-Survival Modeling

Fan Yang, Madelyn Weller, Dimuthu Fernando, Hila Livneh+1

Fowler School of Engineering, Chapman University, Orange, CA 92866 USA · Department of Statistics, Grand Valley State University, Allendale, MI 49401 USA

RUL Estimation Survival Analysis Predictive Maintenance Federated Learning

Open paper AskPDF arXiv

Time-to-event modeling provides a systematic framework for estimating time-dependent failure risk, reliability, and remaining useful life (RUL) from longitudinal condition monitoring data. However, applying these models to distributed prognostics remains challenging because sensor trajectories and failure-time records are often stored across organizations or operational sites and cannot be centrally pooled due to privacy or proprietary constraints. Moreover, the classical Cox proportional hazards model relies on a nonseparable partial likelihood involving global risk sets, making direct optimization difficult under standard federated learning protocols. This paper presents a federated longitudinal-survival modeling framework for collaborative system failure prognostics. The proposed framework combines longitudinal sensor representation learning with a client-separable discrete-time hazard objective, enabling multiple clients to collaboratively train a prognostic model without sharing raw sensor measurements or individual failure records. Time-dependent representations extracted from multivariate sensor histories are used to estimate interval-specific failure hazards, reliability curves, and system RUL. Experiments on the four C-MAPSS turbofan engine degradation subsets under simulated decentralized settings demonstrate that the proposed framework consistently improves prognostic performance over isolated local training while maintaining performance comparable to centralized training across heterogeneous operating conditions and failure modes. These results demonstrate the potential of federated longitudinal-survival modeling for collaborative, data-aware condition monitoring and system failure prognostics.

Abstract

Federated Learning

Explore similar work

Collaborative System Failure Prognostics via Federated Longitudinal-Survival Modeling

Integrating Survival-Based Aging Models with Data-Driven RUL Prognostics

Robust and Personalized Federated Learning for Aircraft-Engine Prognostics under Benign and Adversarial Client Heterogeneity