Stephan Schmidt

This paper evaluates a simulation-driven domain adaptation (SDDA) workflow for few-shot gearbox fault assessment. The study addresses a key limitation of data-driven gearbox fault assessment methods, namely their reliance on large quantities of labelled fault data, which are rarely available in industrial applications. This work investigates whether physics-based simulated source data, a CNN-based data-driven classifier, and MK-MMD-based feature alignment can be effectively combined to assess gearbox faults under severely label-scarce target-domain conditions. The primary contribution of this work is the first systematic verification of a simulation-driven domain adaptation workflow for gearbox fault assessment under industrially relevant few-shot conditions and controlled simulation-to-real discrepancy factors. A comprehensive numerical investigation is conducted to evaluate the effect of the domain discrepancy factors, including model fidelity, sensor location, sensor noise, and physics-based modelling parameter mismatch. The results show that SDDA outperforms the source-only, target-only, and weight fine-tuning baselines across all investigated discrepancy conditions, with the relative performance gain degrading only under extreme sensor noise. The workflow is further evaluated through two experimental case studies using the Prognostics and Health Management datasets, PHM23 and PHM09. These case studies represent relatively well-aligned and substantially mismatched simulation-to-real transfer scenarios, respectively. In both cases, SDDA achieved higher fault detection performance than the considered baselines even with only a few labelled target-domain fault samples. These findings indicate that simulation-based approaches, when coupled with explicit domain alignment, can support data-efficient gearbox fault assessment in industrial environments where real fault data are scarce. Conflict of Interest StatementThe authors declare no conflicts of interest.

Brake pad wear is governed by coupled thermo-mechanical interactions in which degradation alters braking behaviour. This altered braking behaviour, in turn, affects vehicle dynamics, which, in turn, influences future wear evolution. Existing diagnostic and prognostic approaches typically neglect this dynamics–wear coupling, potentially limiting their ability to accurately predict remaining useful life under evolving operating conditions. This work proposes a digital-twin framework for brake wear prognosis that integrates state estimation, parameter adaptation, and dynamics-aware degradation modelling. The approach combines brake pad volume estimation from vehicle operational data, online identification of the brake wear coefficient through inverse modelling, and forward propagation of degradation using a wear-dependent dynamics model. The proposed digital-twin predictive maintenance framework is evaluated using simulated run-to-failure datasets for a mining load-haul-dumper and compared against data-driven and physics-based predictive maintenance models. Results show that the digital-twin approach improves the accuracy and stability of remaining useful life prediction, particularly under anomalous degradation conditions, and achieves earlier convergence to practically useful predictions. These findings demonstrate that accurate brake wear prognosis requires integrating degradation modelling with vehicle dynamics and online parameter updating. The proposed digital twin provides a practical pathway towards more reliable predictive maintenance in systems where degradation and system behaviour are strongly coupled.

Brake pad wear is governed by coupled thermo-mechanical interactions in which degradation alters braking behaviour. This altered braking behaviour, in turn, affects vehicle dynamics, which, in turn, influences future wear evolution. Existing diagnostic and prognostic approaches typically neglect this dynamics-wear coupling, potentially limiting their ability to accurately predict remaining useful life under evolving operating conditions. This work proposes a digital twin framework for brake wear prognosis that integrates state estimation, parameter adaptation, and dynamics-aware degradation modelling. The approach combines brake pad volume estimation from vehicle operational data, online identification of the brake wear coefficient through inverse modelling, and forward propagation of degradation using a wear-dependent dynamics model. The proposed digital twin predicive maintenance framework is evaluated using simulated run-to-failure datasets for a mining load-haul-dumper and compared against data-driven and physics-based predictive maintenance models. Results show that the digital twin approach improves the accuracy and stability of remaining useful life prediction, particularly under anomalous degradation conditions, and achieves earlier convergence to practically useful predictions. These findings demonstrate that accurate brake wear prognosis requires integrating degradation modelling with vehicle dynamics and online parameter updating. The proposed digital twin provides a practical pathway towards more reliable predictive maintenance in systems where degradation and system behaviour are strongly coupled.

Blade Tip Timing (BTT) provides a superior long-term blade monitoring solution compared to strain gauges, yet its adoption in lower value turbomachinery remains constrained by sensor costs. This study explores Inductive Proximity (IP) probes as a cost-effective alternative to eddy current probes. Experimental results demonstrate that the IP probes achieve comparable accuracy in natural frequency estimation while being 40 times more affordable than traditional active eddy current probes. Additionally, IP probes maintain robustness, ensuring reliable vibration parameter extraction via open-source BTT software (version 0.17.0). These findings significantly enhance the accessibility of BTT, enabling budget-conscious turbomachine manufacturers to implement high-precision blade monitoring solutions across a wider range of applications.

The power spectral density (PSD) is a widely used tool for analysing the frequency characteristics of complex vibration signals, playing a central role in fault detection and system monitoring across engineering applications. For high-dimensional datasets, latent variable models (LVMs) based on principal component analysis (PCA) applied to Hankelised time series (TS) have been used recently, primarily for dimensionality reduction. However, PCA-based LVMs remain less established and less intuitive, particularly regarding how they capture and distribute a signal’s frequency-domain information across principal components (PCs). This study explores these internal mechanisms by comparing the PSD of raw, high-dimensional bearing vibration data with the PSD of each individual PC. By examining how power spectral signatures of Hankelised TS are allocated across PCs in the latent space, we gain clearer insight into how the LVM separates and structures distinct frequency features. The aim is to gain a more fundamental and practical understanding of PCA-based LVMs, thereby improving their interpretability and utility in engineering diagnostics.

Optimal filters are designed in vibration-based condition monitoring to enhance weak fault signatures for improved diagnosis. While optimisation-based filter design approaches have matured, their validation has typically focused on final objective values and constraint satisfaction. However, ensuring robust and reliable results requires verifying the correctness of design sensitivities, i.e., the gradients of both objective and constraint functions with respect to design variables, as well as hyperparameter sensitivities. This paper emphasises the importance of confirming and quantifying a filter’s response to varying hyperparameters to ensure it meets design specifications. By rigorously verifying sensitivities, engineers can more confidently deploy optimal filter designs that enhance fault-related features, resulting in more effective fault detection and diagnosis in complex engineering systems.

Latent variable models can effectively determine the condition of essential rotating machinery without needing labelled data. These models analyse vibration data via an unsupervised learning strategy. Temporal preservation is necessary to obtain an informative latent manifold for the fault diagnosis task. In a temporal-preserving context, two approaches exist to develop a condition-monitoring methodology: offline and online. For latent variable models, the available training modes are no different. While many traditional methods use offline training, online training can dynamically adjust the latent manifold, possibly leading to better fault signature extraction from the vibration data. This study explores online training using temporal-preserving latent variable models. Within online training, there are two main methods: one focuses on reconstructing data and the other on interpreting the data components. Both are considered to evaluate how they diagnose faults over time. Using two experimental datasets, the study confirms that models from both training modes can detect changes in machinery health and identify faults even under varying conditions. Importantly, the complementarity of offline and online models is emphasised, reassuring their versatility in fault diagnostics. Understanding the implications of the training approach and the available model formulations is crucial for further research in latent variable model-based fault diagnostics. Conflict of Interest Statement The authors declare no conflicts of interest.