Prognostics of Multisensor Systems with Unknown and Unlabeled Failure Modes via Bayesian Nonparametric Process Mixtures

Kani Fu; Sanduni S Disanayaka Mudiyanselage; Chunli Dai; Minhee Kim

stat.APFeb 22, 2026

Prognostics of Multisensor Systems with Unknown and Unlabeled Failure Modes via Bayesian Nonparametric Process Mixtures

Authors: Kani Fu, Sanduni S Disanayaka Mudiyanselage, Chunli Dai, Minhee Kim

Organizations: Department of Industrial and Systems Engineering, University of Florida, Gainesville, USA · School of Forest, Fisheries, and Geomatics Sciences, University of Florida, Gainesville, USA

Abstract

Modern manufacturing systems often experience multiple and unpredictable failure behaviors, yet most existing prognostic models assume a fixed, known set of failure modes with labeled historical data. This assumption limits the use of digital twins for predictive maintenance, especially in high-mix or adaptive production environments, where new failure modes may emerge, and the failure mode labels may be unavailable. To address these challenges, we propose a novel Bayesian nonparametric framework that unifies a Dirichlet process mixture module for unsupervised failure mode discovery with a neural network-based prognostic module. The key innovation lies in an iterative feedback mechanism to jointly learn two modules. These modules iteratively update one another to dynamically infer, expand, or merge failure modes as new data arrive while providing high prognostic accuracy. Experiments on both simulation and aircraft engine datasets show that the proposed approach performs competitively with or significantly better than existing approaches. It also exhibits robust online adaptation capabilities, making it well-suited for digital-twin-based system health management in complex manufacturing environments.

Figures & tables

Panel of Figure 1 — Figure 1: (a) Stick-breaking process. (b) A sample draw from a DP.

Figure 3: Graphical representation of the proposed framework. Shaded nodes indicate observed variables, while unshaded nodes indicate latent (unobserved) variables or parameters.

Figure 4: Neural network structure of RUL prediction model

Figure 5: Simulation sensor signal plots for the eight sensor signals of three systems (different shapes) per failure mode.

Sensor $s$	$\delta_{s}^{(1)}$	$\delta_{s}^{(2)}$	$\delta_{s}^{(3)}$	$f_{s}(t)$
1	1	0.5	0.9	sine(0.05t)
2	0.1	0.5	0.2	1
3	2	0.01	1	cosine(0.07t)
4	0.001	0.5	0	—
5	0.05	0.5	1	sine(0.1t)
6	0.001	1.5	0.2	sine(0.01t)

Table 1: Sensor-specific parameters

\delta_{s}^{(1)},\delta_{s}^{(2)},\delta_{s}^{(3)}

and temporal functions

f_{s}(t)

used in the simulation data generation

Failure mode $j$	$\delta_{j,s}^{(4)}$
A	[-1, -1, +1, +1, +1, +1, +1, +1]
B	[+1, +1, -1, -1, +1, +1, +1, +1]
C	[+1, +1, +1, +1, -1, -1, +1, +1]
D	[+1, +1, +1, +1, +1, +1, -1, -1]

Table 2: Failure-mode-specific sign parameters

\delta_{j,s}^{(4)}

to control sensor trends under different failure modes

Figure 6: RMSE of RUL prediction and inference number of failure modes over iterations in Simulation dataset

Failure Modes	Similarity Level	Modes Detected	NMI	RMSE
2-Modes	Low ( $-1.00$ )	2 modes	1.00	12.94
2-Modes	High ( $0.75$ )	2 modes	1.00	16.87
3-Modes	Low ( $0.17$ )	3 modes	1.00	10.58
3-Modes	High ( $0.33$ )	3 modes	1.00	11.07
4-Modes	Low ( $0.17$ )	4 modes	1.00	19.34
4-Modes	High ( $0.33$ )	4 modes	1.00	21.53

Table 3: RUL prediction performance under different failure mode complexity and cosine similarities

Failure Modes	Method	Mode Detected	NMI	RMSE
2-Modes	DPMM-joint	1 Mode	0.00	64.91
	DPMM-RUL (ELBO)	1 Mode	0.00	15.50
	DPMM-RUL (RUL loss)	3 Modes	0.82	14.92
	DPMM-RUL ( $\mathcal{J}$ score)	2 Modes	1.00	12.94
3-Modes	DPMM-joint	1 Mode	0.00	23.34
	DPMM-RUL (ELBO)	1 Mode	0.00	15.69

Table 4: RUL prediction performance comparison across different performance scores

Failure Modes	$\mathbf{\omega}$	Similarity	Mode Detected	NMI	RMSE
3-Modes	$1\times 10^{-4}$	Low	3 modes	1.0	13.58
	$1\times 10^{-3}$	Low	3 modes	1.0	11.07
	$1\times 10^{-2}$	Low	3 modes	1.0	16.59
	$1\times 10^{-1}$	Low	3 modes	1.0	11.46
3-Modes	$1\times 10^{-4}$	High	3 modes	1.0	11.52
	$1\times 10^{-3}$	High	3 modes	1.0	10.58

Table 5: Sensitivity analysis of the trade-off coefficient

\omega

under varying failure mode complexities

Figure 7: Number of inferred failure modes over iterations. The black dotted line shows when the new failure mode appears.

Method	Failure mode known		NMI	RMSE (after shift)
Method	Label	Number	NMI	RMSE (after shift)
Li et al. (2023)	$\checkmark$	$\checkmark$	0.00	60.24
Wang et al. (2024a)	$\checkmark$	$\checkmark$	0.73	63.93
Fu et al. (2025)	$\times$	$\checkmark$	–	18.02
DPMM-joint	$\times$	$\times$	0.00	19.82
DPMM-RUL (ELBO)	$\times$	$\times$	0.00	12.51

Table 6: RUL prediction performance under increasing uncertainty in failure-mode information when the number of modes changes from 2 to 3

Symbol	Description	Units
T2	Total temperature at fan inlet	∘ R
T24	Total temperature at LPC outlet	∘ R
T30	Total temperature at HPC outlet	∘ R
T50	T50 Total temperature at LPT outlet	∘ R
P2	Pressure at fan inlet	psia
P15	Total pressure in bypass-duct	psia

Table 7: Detailed description of multiple sensors in CMAPSS dataset

Panel of Figure 8 — Figure 8: RMSE of RUL prediction and inferred number of failure modes over iterations on FD0003 with (a) truncation and padding data preprocessing; and (b) temporal warping preprocessing.

Figure 9: Comparison between inferred and true failure modes in three representative sensors

Method	Failure mode known		NMI	RMSE	Training time(s)	Testing time(s)
Method	Label	Number	NMI	RMSE	Training time(s)	Testing time(s)
Li et al. (2023)	$\checkmark$	$\checkmark$	0.66	17.67	947.35	2.88
Wang et al. (2024a)	$\checkmark$	$\checkmark$	0.80	19.80	1891.49	0.06
Fu et al. (2025)	$\times$	$\checkmark$	-	24.19	6013.00	4.03
DPMM-joint	$\times$	$\times$	0.00	41.11	1335.41	0.73
DPMM-RUL (ELBO)	$\times$	$\times$	0.00	42.60	3789.42	0.06

Table 8: RUL prediction performance comparison on C-MAPSS FD003 under different availability of failure-mode information

Dirichlet Process Mixture Models

See all 8 papers

Prognostics for Autonomous Deep-Space Habitat Health Management under Multiple Unknown Failure Modes
Nov 19, 2024Benjamin Peters, Ayush Mohanty, Xiaolei Fang +2RUL Estimation Fault Diagnosis
University of Texas at Rio Grande Valley, Edinburg, TX 78539, USA · Georgia Institute of Technology, Atlanta, GA 30332, USA · North Carolina State University, Raleigh, NC 27607, USA +1
A Continual Validation, Updating, and Decision-Making Framework for Self-Adaptive Digital Twins via Robust Model Predictive Control: A Case Study in Additive Manufacturing
Jul 20, 2026Yi-Ping Chen, Ying-Kuan Tsai, Vispi Karkaria +3Neural Surrogate Modeling Concept Drift
Department of Mechanical Engineering, Northwestern University · Department of Industrial Engineering and Management Science, Northwestern University
Fast Bayesian equipment condition monitoring via simulation based inference: applications to heat exchanger health
Apr 22, 2026Peter Collett, Alexander Johannes Stasik, Simone Casolo +1Simulation-Based Inference Efficient Inference
Cognite AS, Oslo, Norway · Department of Data Science, Norwegian University of Life Sciences, Ås, Norway · Department for Mathematics and Cybernetics, SINTEF Digital, Oslo, Norway

Nov 19, 2024J/K move · Enter open · S save

Prognostics for Autonomous Deep-Space Habitat Health Management under Multiple Unknown Failure Modes

Benjamin Peters, Ayush Mohanty, Xiaolei Fang, Stephen K. Robinson+1

University of Texas at Rio Grande Valley, Edinburg, TX 78539, USA · Georgia Institute of Technology, Atlanta, GA 30332, USA · North Carolina State University, Raleigh, NC 27607, USA+1

RUL Estimation Fault Diagnosis

Open paper AskPDF arXiv

Deep-space habitats (DSHs) are safety-critical systems that must operate autonomously for long periods, often beyond the reach of ground-based maintenance or expert intervention. Monitoring system health and anticipating failures are therefore essential. Prognostics based on remaining useful life (RUL) prediction support this goal by estimating how long a subsystem can operate before failure. Critical DSH subsystems, including environmental control and life support, power generation, and thermal control, are monitored by many sensors and can degrade through multiple failure modes. These failure modes are often unknown, and informative sensors may vary across modes, making accurate RUL prediction challenging when historical failure data are unlabeled. We propose an unsupervised prognostics framework for RUL prediction that jointly identifies latent failure modes and selects informative sensors using unlabeled run-to-failure data. The framework consists of two phases: an offline phase, where system failure times are modeled using a mixture of Gaussian regressions and an Expectation-Maximization algorithm to cluster degradation trajectories and select mode-specific sensors, and an online phase for real-time diagnosis and RUL prediction using low-dimensional features and a weighted functional regression model. The approach is validated on simulated DSH telemetry data and the NASA C-MAPSS benchmark, demonstrating its ability to identify unknown failure modes, select mode-specific informative sensors, and accurately predict RUL.

Abstract

Dirichlet Process Mixture Models

Explore similar work

Prognostics for Autonomous Deep-Space Habitat Health Management under Multiple Unknown Failure Modes

A Continual Validation, Updating, and Decision-Making Framework for Self-Adaptive Digital Twins via Robust Model Predictive Control: A Case Study in Additive Manufacturing

Fast Bayesian equipment condition monitoring via simulation based inference: applications to heat exchanger health