A Hybrid Digital Twin Framework for Carbon Brush Degradation State Estimation and Remaining Useful Life Prediction in PMDC Motors = 브러시드 DC 모터의 카본 브러시 열화 상태 추정 및 잔여 수명 예측을 위한 하이브리드 디지털 트윈 프레임워크

다국어 초록 (Multilingual Abstract)

This research presents a hybrid digital twin framework for carbon brush degradation estimation and remaining useful life (RUL) prediction in permanent magnet DC (PMDC) motors. The study addresses a major limitation in carbon brush maintenance: brush length cannot be directly measured during operation because inspection requires stopping and disassembling the motor, resulting in operational interruption and potential safety risks. Consequently, degradation must be inferred indirectly from measurable operational signals.
To address this limitation, a physics-based Simulink motor model was integrated with a data-driven vibration analysis pipeline implemented in Python. High-frequency vibration signals acquired from a commutation-side accelerometer were processed to extract degradation-sensitive features from both the time and frequency domains. Following feature evaluation and redundancy reduction, a fused health indicator was constructed and calibrated into a monotonic degradation coordinate using isotonic regression and spline smoothing.
An Unscented Kalman Filter (UKF) was subsequently employed to recursively estimate degradation severity and degradation rate by combining the physics-based degradation model with the measurement-derived observations. The estimated degradation states were coupled to the motor electrical parameters through a degradation map linking brush wear to increasing contact resistance. Using the estimated degradation state and degradation rate, the digital twin generated RUL predictions with associated uncertainty bounds throughout operation.
Experimental validation demonstrated stable degradation tracking throughout the wear progression. The framework achieved an RUL mean absolute error of 4.05 hours and an RMSE of 7.60 hours over the degradation trajectory, while reconstructed brush lengths remained physically consistent with the measured terminal wear condition. The results demonstrate that hybrid physics-informed and data-driven architectures provide a viable approach for degradation estimation in systems where the dominant damage mechanism cannot be directly measured during operation.

목차 (Table of Contents)

• Chapter 1: Introduction 1
• 1.1 Introduction 1
• 1.1.1 Evolution of Maintenance Strategies 2
• 1.1.2 PHM: Framework and functions 3
• 1.1.3 Limitations of current PHM and Motivation for DT Integration 3
• 1.1.4 Industrial Context: Industry 4.0 and Industry 5.0 4
• 1.1.5 Research Focus: Carbon Brush Degradation in PMDC Motors 6
• 1.2 Study Objectives and Problem Statement 7
• 1.3 Constructive Contributions 8
• 1.4 Dissertation Roadmap 9
• Chapter 2: Literature Review 10
• 2.1 From Condition monitoring to PHM: The state Centered Paradigm 10
• 2.2 RUL estimation and their structural evolution 13
• 2.2.1 Direct prediction Methods 13
• 2.2.2 State-Based Methods 14
• 2.2.3 Hybrid and DT-Oriented Methods 16
• 2.3 Health State Construction, Observability and Uncertainty 18
• 2.4 PHM In Electric Motors: Advances, Research Bias and unresolved challenges 21
• Chapter 3: Methodology and Experimetal Testbed 24
• 3.1 Overview 24
• 3.2 The Carbon Brush as Prognostic Target 26
• 3.2.1 Operational Role of the Brush-Commutator Interface 26
• 3.2.2 Observability Constraints in Brush Wear Monitoring 28
• 3.3 Motor Model and Electromechanical Dynamics 29
• 3.3.1 Electromechanical Dynamics of the PMDC Motor 29
• 3.3.2 Brush-Commutator Interaction and Characteristic Excitation Frequen-cies 32
• 3.4 Experimental Testbed Description 34
• 3.4.1 Experimental Operating conditions 35
• 3.4.2 Instrumentation and Sensor Placement 36
• 3.5 Experimental Design 37
• 3.5.1 Dataset Acquisition 38
• 3.5.2 Brush-Length Measurements and Degradation Reference 38
• 3.5.3 Dataset Role Within the DT Framework 39
• 3.6 Signal Processing and HI Construction 39
• 3.6.1 Signal Preprocessing 40
• 3.6.2 Feature Extraction 40
• 3.6.3 Feature evaluation and selection 41
• 3.6.4 HI Fusion 43
• 3.6.5 Calibration to a Degradation Coordinate 44
• 3.6.6 Measurement Noise Estimation 45
• 3.7 Simulink DT 46
• 3.7.1 State Definition 47
• 3.7.2 Process Model 47
• 3.7.3 Observation Model 49
• 3.7.4 UKF 50
• 3.7.5 Parameter Coupling 51
• 3.7.6 RUL Projection 51
• Chapter 4: Results 53
• 4.1 Data Ingestion and Validation 53
• 4.2 Frequency-Domain Analysis 54
• 4.3 Signal Conditioning 56
• 4.4 Feature Extraction Results 57
• 4.5 Feature Selection and Redundancy Reduction 59
• 4.6 HI Construction 62
• 4.7 Calibration and Export 65
• 4.8 Simulink Digital Twin Implementation 67
• 4.8.1 Motor Model Verification 67
• 4.8.2 UKF Configuration 68
• 4.8.3 Degradation-to-Parameter Coupling 69
• 4.8.4 UKF Estimation Results 71
• 4.8.5 Observation Model Consistency and Innovation Behaviour 72
• 4.8.6 Estimator Confidence and Posterior Covariance Evolution 74
• 4.8.7 Brush Length Reconstruction and Physical Validation 76
• 4.8.8 RUL Prediction with Uncertainty Bounds 78
• 4.8.9 Prognostic Accuracy and Error Characterisation 79
• Chapter 5: Discussion and Conclusion 82
• 5.1 Observaility of Carbon Brush Degradation 82
• 5.2 Behaviour of the Recursive Estimator 83
• 5.3 Physical consistency of the DT 85
• 5.4 Prognostic Capability and Uncertainty Behaviour 87
• 5.5 Limitations of the Present Study 88
• 5.6 Future Work 89
• References 92