Annual Conference of the PHM Society

FE-Based Data Generation for Structural Integrity Verification with Neural Networks

Maximilian Posner, Martin Dazer — Mon, 28 Sep 2026 00:00:00 +0000

The application of deep learning within Prognostics and Health Management (PHM) frameworks offers immense potential for real-time structural health monitoring. However, in high-consequence applications such as high-velocity collisions, algorithmic deployment is severely bottlenecked by a fundamental scarcity of training data. Because destructive physical testing is prohibitively expensive and real-world incident telemetry is rare, acquiring the massive, diverse datasets required to prevent neural network overfitting remains a critical operational challenge. To overcome this data scarcity, this paper presents a scalable methodology that transforms high-fidelity Finite Element (FE) analysis into an automated Physics-of-Failure "data engine". Using the safety-critical underrun guard of a high-speed intercity train as a primary case study, constructing a complete Digital Thread that bridges quantitative continuum mechanics with qualitative human operational logic.

To prevent the curse of dimensionality during automated data generation, this framework implements rigorous physical and algorithmic boundaries. Complex multi-body impact kinematics are mathematically distilled, and six non-linear material parameters are algorithmically condensed into a singular "toughness" variable to honor the physical inverse correlation between material strength and ductility. Utilizing a low-discrepancy Sobol sequence, the simulation parameters, are sampled uniformly with 256 transient-dynamic simulations, ensuring the highly sensitive boundary transitions of structural failure are densely captured without statistical clustering.

To simulate real-world hardware constraints, a virtual sensor network of accelerometers and strain gauges is deployed. Multi-axial strain tensors are mathematically reduced to Von Mises scalar invariants, optimizing the AI input layer and preventing feature bloat. Concurrently, standardized multi-angle visual renderings of the deformed FE mesh are generated, enabling domain experts to label the physical damage into discrete, actionable operation and maintenance directives (continue operation, reduced operation, immediate operational stop and reuse, repair, exchange).

Following the experimental validation of the digital twin against physical destructive laboratory tests, the resulting synthetic dataset was utilized to train a baseline classification neural network. Despite the severely constrained sample size ($n = 256$), the diagnostic AI achieved an outstanding predictive accuracy of 82 \%. This confirms that mathematically optimized, physics-grounded data generation can successfully bridge the data gap in modern PHM frameworks, providing a robust pathway for real-time diagnostic deployment.

Self-Supervised Representation Learning for Remaining Useful Life Prediction:

Mohammad Badfar, Murat Yildirim, Ratna Babu Chinnam — Mon, 28 Sep 2026 00:00:00 +0000

Remaining useful life (RUL) prediction is fundamentally constrained by the scarcity of labeled run-to-failure data, motivating growing use of self-supervised learning (SSL) to exploit abundant unlabeled condition monitoring data. However, the literature lacks a systematic comparison of the loss functions underlying these SSL methods, as existing surveys organize the field by architecture rather than by the training objective itself. This paper reviews self-supervised loss functions for RUL prediction, spanning temporal-order, contrastive, metric-learning, multi-view, and meta-learning objectives, summarizing each family's assumptions, requirements, and limitations. To ground this review empirically, we present a framework pre-training a Transformer encoder with a triplet margin loss and evaluating it under few-shot fine-tuning on the C-MAPSS turbofan benchmark. Across 2 to 50 labeled fine-tuning sequences, the pre-trained model consistently outperforms an identically structured supervised baseline, with the gap narrowing as labeled data becomes abundant, confirming that SSL pre-training is most valuable precisely in the label-scarce regime.

Toward Statistical Risk Analysis for Aerospace Systems with AI/ML Components

Yuning He, Konstantin Dmitriev — Mon, 28 Sep 2026 00:00:00 +0000

Aerospace systems increasingly rely on machine learning (ML) and AI components. Such safety-critical applications require careful verification and validation (V&V), and certification. A central task in safety certification is risk assessment: for a large number of off-nominal failure conditions, it must be demonstrated that they do not pose undue risk during operation. Events with major, hazardous, or catastrophic consequences must only occur extremely infrequently.

This paper addresses the statistical modeling and analysis of rare events in AI/ML-based components at both the systemand component-level within the SYSAI analysis framework. Here, we focus on typical perception-related functional tasks that are found in unmanned aerial vehicles (UAVs) or supporting aircraft operations, specifically an Autonomous Centerline Tracking (ACT) system and an Autonomous Emergency Braking System (AEBS). We provide an overview of the SYSAI statistical learning and analysis framework and discuss challenges for the analysis of failure modes and rare events in systems with AI/ML components. We describe how SYSAI can efficiently generate and evaluate scenarios that are likely to contain rare events. These experiments support the estimation of reliable probability distributions for such events and the analysis of temporal dependencies between individual input images.

Comparative Evaluation of Multimodal Fusion Architectures for Cyber Physical System Anomaly Classification

Mon, 28 Sep 2026 00:00:00 +0000

Smart manufacturing systems generate heterogeneous data from programmable logic controllers (PLCs), industrial networks, and system logs, providing complementary information for cyber-physical system health monitoring. However, few studies have systematically compared multimodal fusion architectures for industrial anomaly classification. This paper evaluates four architectures, Fusion MLP, Fusion 1D CNN, Fusion CNN LSTM, and Cross Modal Attention Fusion, using synchronized PLC telemetry, network traffic, and system log data under an identical experimental protocol. Unimodal and classical machine learning baselines are also evaluated to assess the contribution of multimodal fusion. Results show that all fusion architectures achieve near-perfect classification performance. However, unimodal and temporal evaluations reveal that the benchmark is highly separable, with PLC telemetry alone achieving 99.93% accuracy and 99.68% macro F1 under temporal evaluation. Confusion matrices, t-SNE visualization, and attention allocation analysis further examine model behavior and modality contributions. The findings show that multimodal fusion effectively integrates heterogeneous industrial data, while increased architectural complexity does not necessarily improve classification performance when individual modalities contain highly discriminative signals. These results highlight the importance of unimodal baselines, temporal robustness evaluation, and interpretable multimodal analysis in assessing industrial anomaly detection systems.

Cluster-Based Permutation Tests for Automated Identification of Critical Frequency Bands in Vibration Data

Neil Eklund, Adam Wechsler, Colin Dingley, Sherwood Polter, Khai Van — Mon, 28 Sep 2026 00:00:00 +0000

Condition-based maintenance plus for naval applications requires robust embedded health management systems. However, extracting reliable diagnostic features from reciprocating machinery in harsh maritime environments is hindered by severe acoustic interference and non-stationary dynamics. Traditional point-wise statistical methods, such as analysis of variance, suffer from the multiple comparisons problem and select redundant, highly correlated frequency bins. Conversely, unsupervised dimensionality reduction techniques like principle component analysis are susceptible to broadband noise and lack physical interpretability.
To address these limitations, this paper presents a hierarchical spatial-spectral feature selection framework based on cluster based permutation tests (CBPT). By evaluating multivariate spectral adjacency and incorporating stratified block bootstrap stability selection, the method isolates deterministic kinematic resonances. This non-parametric approach rejects stochastic background noise and mitigates the large-sample paradox without requiring manual frequency band selection.
The framework was validated using multi-channel vibration data from a reciprocating electromechanical testbed subjected to working fluid leakage, intake restriction, and thermal management degradation. To assess generalization, models were evaluated using a chronological train-test split with progressive levels of additive white Gaussian noise added to the test sets. The CBPT-derived feature space maintains a macro F1-score above 0.90 down to a 5 dB signal-to-noise ratio, outperforming standard univariate and broadband extraction baselines under acoustic interference. This methodology yields an independent, low-dimensional feature set suitable for autonomous, embedded prognostic architectures.

The “Last Mile” of PHM:

Simon Trussler, Ralph Poole, Ken Percell, Frank Zahiri — Mon, 28 Sep 2026 00:00:00 +0000

To accelerate the exhaustive process of military airworthiness certification, this paper introduces a deterministic, Natural Language Processing driven workflow that establishes a Verifiable Claim Architecture (VCA). While PHM strategies successfully detect anomalies and predict subsystem failures, mapping these low-level engineering findings to high-level regulatory mandates remains a massive documentation bottleneck. The proposed VCA automates evidence discovery by integrating multi-format layout parsing and reproducible Boolean filtering to establish a highly structured, auditable evidence layer. Large Language Models are then deployed strictly as constrained text-synthesis engines, operating within isolated context windows to populate formal compliance matrices.

To fundamentally eliminate textual hallucination and abstractive drift, a provenance audit layer anchors every generated narrative block to an unalterable source document metadata tag. Furthermore, the architecture integrates system safety ontologies to automatically map engineering deviations to standardized Risk Assessment Codes (RAC) and operational flight restrictions in accordance with MIL-STD-882E. Evaluated by the Air Force Life Cycle Management Center against 7,346 pages of compliance artifacts across multiple aircraft platforms, the framework demonstrated a highly conservative safety posture, properly deferring policy exceptions to human engineers. By reducing expert administrative labor hours by an estimated 50% and compressing certification schedules from 18–36 months down to 6–12 months, this domain-agnostic framework effectively streamlines the compliance lifecycle, ensuring human experts remain focused entirely on authoritative, high-consequence risk assessment.

A Hybrid Physics and Machine Learning Digital Twin for Remaining Useful Life Prediction of SiC Power Modules in Traction Inverters

Kooros Moabber, Hamed Talebi, Mahmoud Rahat — Mon, 28 Sep 2026 00:00:00 +0000

Silicon Carbide (SiC) power modules are the most failure-prone component in automotive traction inverters, responsible for the large majority of reported inverter faults. There are two established strategies for estimating power module lifetime: a physics-based approach and a data-driven approach. Each strategy has complementary weaknesses: the physics-based approach cannot adapt to real-world degradation and is computationally heavy while pure ML approaches require large high-quality datasets, extrapolate poorly outside their training range and can produce physically unrealistic and hard-to-justify predictions. This paper proposes and evaluates a hybrid physics and ML digital twin architecture that corrects a physics-based RUL estimate by using a machine-learned residual model trained on power-cycling data from 20 SiC power modules. Input features (on-state voltage, junction temperature swing, junction temperature, thermal resistance, accumulated fatigue damage and their moving averages) are standardized and reduced via Principal Component Analysis (PCA) to a single health indicator, which together with the physics-based damage estimate feeds a residual RUL model. Six candidate regression architectures, a multilayer perceptron (MLP), Extreme Gradient Boosting (XGBoost), linear regression, ridge regression, lasso regression and a quadratic response-surface model (Box-Behnken) are compared by using 5-fold cross-validation. XGBoost and the quadratic model achieve the best trade-off between accuracy (mean R^2 ~ 0.87) and embedded implementability although the MLP achieves the highest raw accuracy (R^2 = 0.94) at higher computational cost. We further demonstrate online RUL tracking and residual-based anomaly detection under two field-data scenarios (physics-pessimistic and physics-optimistic) and discuss a path toward embedded deployment and further hybrid blending strategies.

Part Analytics using Digital Threads

Miriam Alvarez-Pintor, Santosh Bhatt — Mon, 28 Sep 2026 00:00:00 +0000

Determining whether a component is Beyond Economical Repair (BER) depends on multiple technical and business factors, including repair cost thresholds, replacement/market value, component availability and lead time, persistency of faults (recurrence after repair), and the time required to restore operational readiness. This work evaluates a predictive, data-driven approach that uses a digital thread tying serialized parts to their host aircraft and leverages aircraft utilization forecasts (e.g., from FlightRadar24) to generate flight-hour-based usage forecasts. By combining usage forecasts with part-level reliability models and historical failure data, we produce one-year risk forecasts for part failure (remaining useful life / failure probability).

Enriching the digital thread with part attributes (repair cost, new-purchase cost, repair cycle time, vendor lead time) and part-family trend analytics enables a cost–benefit analysis of repair versus replacement that explicitly considers total cost of ownership and operational downtime. Applying configurable business rules and inventory state (on-hand, pipeline, critical spares) allows the system to recommend actions (repair, replace, or retire excess inventory) that minimize expected lifecycle costs and mission-impacting downtime. This paper presents the approach that supports proactive supply-chain and maintenance decisions, improves fleet availability, and identifies systemic part-family issues that may warrant engineering or supplier corrective action.

Resource-Aware Model Comparison for DASHlink Multi-Class Flight-Anomaly Screening

Keanan Milton, Ge Zhou — Mon, 28 Sep 2026 00:00:00 +0000

Flight-data prognostics and health management models have to do more than rank anomalous windows correctly. For future edge or onboard validation, a screening model also has to be compact, fast to score, false-alarm aware, and interpretable enough that its decision statistic can be trusted. Autoencoders are attractive in this setting because they can be trained from nominal data, but a single scalar reconstruction error may be too coarse to separate operationally different non-nominal approach behaviors. We tested that tradeoff on the NASA DASHlink multi-class flight-anomaly window benchmark using a fixed split, nominal-only threshold calibration, and a held-out test set. Each sample is a 160-second, 20-variable window labeled as Nominal, Speed High, Path High, or Flaps Late Setting. We treat the primary task as nominal-versus-non-nominal screening, while preserving class-wise recall so that the rare classes are not hidden by aggregate recall. The comparison includes raw-summary classifiers, classical unsupervised baselines, scalar convolutional variational autoencoder (CVAE) reconstruction scores, CVAE residual and latent representations, and raw-plus-CVAE hybrid models. Thresholded metrics use a nominal-calibrated p95 operating point; resource proxies include total scoring time and serialized pipeline size. Scalar CVAE reconstruction error was weak as a standalone detector, with AP 0.179 and worst-class recall 0.051. Residual and latent CVAE features retained substantially more class-discriminative information, and the strongest raw-plus-CVAE hybrid reached AP 0.935, recall 0.945, and worst-class recall 0.894 at 0.158 ms/window. Full-20 raw-summary HGB remained the simpler high-performing baseline. The practical result is that compact feature-based models are strong screening baselines, while reconstruction models are most useful here as auxiliary representation generators rather than standalone scalar anomaly detectors.

A Cyber-Physical Testbed for Hierarchical Fault Isolation in Bilateral Systems: Multi-Domain Fusion with Attention-Derived Laterality Attribution

Jagadeesh Harinarayanan — Mon, 28 Sep 2026 00:00:00 +0000

Prognostics and Health Management (PHM) for complex cyber-physical systems is constrained by a fundamental scarcity of labeled fault data, as catastrophic failures are operationally irreproducible and ethically impossible to induce at scale. To bridge this gap, this study utilizes a dual-engine aerospace platform as a high-fidelity case study, detailing a reproducible cyber-physical testbed leveraging the DCS World Su-25T simulation environment for hierarchical fault isolation. A robust real-time telemetry architecture extracts 25 sensor channels at a 30 Hz scheduler frequency via Apache Kafka and PostgreSQL, facilitating programmatic fault injection across a structured 3×3 altitude-speed grid (1,000 ft, 8,000 ft, and 20,000 ft altitude bands; 350, 400, and 450 kts TAS). The resulting dataset comprises 369 labeled missions spanning four distinct thermodynamic states: left-engine fire, right-engine fire, dual-engine fire, and a nominal baseline. Under a validated injection protocol with randomized onset (T+60s to T+80s), two research questions are addressed. RQ1 evaluates multi-domain sensor fusion through an XGBoost framework trained on a 789-feature vector encompassing time-domain statistics and cross-channel asymmetry indicators. RQ2 explores temporal fault laterality via a PyTorch Transformer encoder processing 450-timestep post-fault sequences. Model interpretability is achieved through SHAP domain contribution analysis and Attention-derived Domain Activation Profiles (DAPs). The XGBoost pipeline demonstrates exceptional diagnostic performance (Macro F1 = 0.9914), while the Transformer achieves Macro F1 = 0.9784 for laterality identification, maintaining robustness via leave-2-out cross-validation (mean F1 = 0.975). A novel methodological finding reveals that the Transformer identifies failure lateralization by attending to stable healthy-side channels, suggesting an information-theoretic entropy differential mechanism in bilateral systems.

Simulation-to-real Unsupervised Domain Adaptation Gear Diagnostics Based on Entropy

Mon, 28 Sep 2026 00:00:00 +0000

Gearbox failures are a major contributor to the levelized cost of electricity of wind turbines, owing to their long repair time and large share of total turbine downtime. Condition monitoring enables early fault detection, reducing unplanned downtime and the associated economic impact, while digital twins provide virtual representations of physical assets that support advanced diagnostics. This work couples the two: a phenomenological gear vibration model generates a synthetic dataset spanning healthy operation and several gear pitting severities, and healthy measurements from the real asset are used to estimate a transfer function, through cepstrum liftering, that is applied to the synthetic signals so that they better reproduce the experimental response. By stochastically varying the model parameters within prescribed ranges, a labeled synthetic dataset is produced and used to train a deep learning unsupervised domain adaptation (UDA)networkthatcouples adversarial training with entropy-based sample weighting and an information maximization loss, reducing the mismatch between simulated and measured data and improving diagnostic robustness on real measurements. The framework is validated on vibration data from an in-house back-to-back, single-stage helical gear test rig, in which pitting was artificially introduced on the tooth flank. Against state-of-the-art (SOTA) adversarial methods and a source-only baseline, the proposed approach more accurately distinguishes healthy operation from multiple pitting severity levels, supporting its potential for diagnosis on real applications.

Disentangling Prognostic Observability from Policy Stochasticity in PHM-Aware Reinforcement Learning for Semiconductor Fab Dispatch

Sean Mondesire, Tori Wright, Bulent Soykan — Mon, 28 Sep 2026 00:00:00 +0000

High-mix low-volume (HMLV) semiconductor manufacturing is a demanding proving ground for prognostics and health management (PHM): frequent product changeovers, re-entrant routes, and heavy tool-qualification overhead amplify the operational cost of unexpected failures, so healthaware dispatch justify its sensing investment. This work investigates that promise in a reliability-aware HMLV fab simulator using a 2×2 ablation of two deep reinforcement learning proximal policy optimization (PPO) dispatchers (PHMaware vs. dispatch-only) and two deployment modes (deterministic argmax vs. stochastic sampling), all sharing reward, training budget, and seeds. Across 10 seeds and 10 heldout HMLV order sets, deployment mode dominated: stochastic PHM-aware PPO cut mean makespan from 48,939.9 to 24,387.2 minutes and lifted OEE from 0.707 to 0.829 (all FDR-significant), while PHM-aware and dispatch-only PPO were statistically indistinguishable under matched deployment. The result sharpens the PHM-awareness case for HMLV fabs: prognostic value is real but contingent on how learned policies turn information into action.

Physics-Informed Operator Learning for Real-Time Battery State Estimation and Health Monitoring

Alexander Harej, Marco Giglio, Francesco Cadini — Mon, 28 Sep 2026 00:00:00 +0000

The increasing demand for reliable and safe lithium-ion (Li-ion) battery packs in the automotive sector has intensified the need for advanced diagnostics and prognostics capabilities within next-generation Battery Management Systems (BMSs). A key BMS function is the estimation of the battery State of Charge (SOC) and State of Health (SOH), typically achieved through Kalman-based observers that require an explicit state-space representation of the cell. Equivalent Circuit Models (ECMs), while computationally efficient, are fundamentally empirical constructs that offer limited physical insight and poor generalization outside their identification range. Electrochemical Models (EMs), by contrast, capture the underlying reaction kinetics, mass transport, and diffusion processes governing cell behavior, enabling higher fidelity across a broader range of operating conditions at the cost of computational complexity that has, so far, precluded their real-time deployment within BMS observers. This work addresses this barrier by relying on a physicsinformed surrogate of the Single Particle Model (SPM), based on a Multiple-Input Operator Network (MIONet) that directly encodes the SPM governing equations into a compact, real-time-executable state-space representation. The resulting EM-based observer is embedded within an Unscented Kalman Filter (UKF) for the joint estimation of SOC and SOH, where SOH is represented by the Loss of Lithium Inventory (LLI, in moles) augmented to the SPM concentration state. The framework is validated on a high-fidelity Doyle-Fuller-Newman (DFN) plant including electrochemically resolved Solid Electrolyte Interphase (SEI) growth, lithium plating, and loss of active material, simulated in PyBaMM. Six case studies covering Constant Current-Constant Voltage (CC-CV) charges followed by CC and Dynamic Stress Test (DST) discharges at 10◦C, 25◦C, and 40◦C are considered, in which the cell is cycled to its end-of-life (EOL, 80% of initial capacity). Quantitative comparisons against an ECM-based observer with an identical joint-UKF structure show that the EM-based observer consistently delivers lower SOC and SOH Root Mean Square Error (RMSE) and substantially better-calibrated confidence intervals (measured via the Coverage Width-based Criterion, CWC). The largest advantage is observed under highly dynamic load profiles, establishing the proposed approach as a viable pathway toward high-fidelity, real-time diagnosis in next-generation BMSs.

Explainable Temporal Attribution for Time-Series-Based Quality Prediction in Injection Molding

Yulin Wang, Elke Deckers, Konstantinos Gryllias — Mon, 28 Sep 2026 00:00:00 +0000

Time-series sensor signals collected during injection molding
contain rich information about process dynamics and final
part quality. However, quality prediction models based on
high-dimensional temporal signals are often difficult to in
terpret, and direct attribution in the original signal domain
may produce unstable explanations. This paper proposes an
explainable temporal attribution framework for time-series
based quality prediction in injection molding. The framework
first learns compact latent representations of multivariate pro
cess signals using a one-dimensional convolutional autoen
coder, and then trains a latent-space prediction model for
defect classification or dimensional quality regression. To re
cover process-level interpretability, SHapley Additive exPlana
tions (SHAP) are computed in the latent space and propagated
back to the original temporal signal domain through the de
coder Jacobian. The proposed framework is evaluated on two
injection molding datasets. The results show that the learned
latent representations preserve dominant temporal process
characteristics and provide useful information for downstream
quality prediction. The temporal attribution results reveal that
pressure-related signals and the filling-to-packing transition
are especially important for quality prediction, while phase
aware analysis indicates that packing and cooling stages play
dominant roles in sink mark prediction. Robustness analysis
further demonstrates that the decoder-based temporal attribu
tion framework provides more stable and consistent explana
tions than direct SHAP analysis in the original signal domain.
The proposed method offers an interpretable way to connect
latent predictive features with physically meaningful mold
ing process dynamics, supporting process understanding and
quality-related decision making.

A Knowledge-Graph-Guided Diagnostic Pipeline for Root Cause Analysis of Equipment Failures in Nuclear Power Plant

Diego Mandelli, Congjian Wang — Mon, 28 Sep 2026 00:00:00 +0000

The root cause analysis of complex equipment failures in nuclear power plants requires the integration of heterogeneous data sources into traceable causal explanations. It is a process that can be time-consuming, expert-dependent, and difficult to audit across plant lifetime. This paper presents a structured diagnostic reasoning framework designed to copilot system engineers to perform root cause analyses, improving analysis speed, consistency, and traceability while preserving full engineering interpretability. We have formally decomposed a causal reasoning architecture that integrates five independent but interacting reasoning dimensions (i.e., structural, temporal, evidence-based, governance, and historical) allowing each dimension of the diagnostic process to be explicitly validated. The framework integrates several heterogeneous inputs such as telemetry signals, plant documentation, system architecture models, and operational context within a unified knowledge representation. Structural reasoning constrains the causal search space by identifying physically plausible failure modes based on system connectivity and component dependencies. Temporal reasoning characterizes the relationships between observed anomalies and the main event using interval-based representations to distinguish candidate causes from coincident or downstream effects. Evidence-based reasoning retrieves and evaluates relevant historical records, classifying them as supporting, contradicting, or contextual with respect to each candidate hypothesis. Candidate causes are generated and evaluated across multiple dimensions, including structural plausibility, temporal consistency, telemetry alignment, and documentary evidence strength, with explicit tracking of uncertainty and conflicting evidence. A synthesis stage produces an assessment that includes ranked hypotheses, supporting evidence with traceable references, identified uncertainties, and recommended follow-up actions. A key design principle of the framework is that causal conclusions are only promoted when multiple independent sources of evidence align, reducing the risk of premature convergence on incomplete or biased explanations. The framework is evaluated through mechanismlevel validation across three representative nuclear plant failure scenarios covering condenser degradation, reactor trip temporal gating, and surveillance check valve leakthrough.

A Cross-Phase Prognostic Framework for Emergent Work Risk Assessment and Schedule Decision Support in Nuclear Refueling Outages

Diego Mandelli, Congjian Wang — Mon, 28 Sep 2026 00:00:00 +0000

Refueling outages are among the most schedule-critical phases in a nuclear plant’s operating life. The premise of this paper is that outages are governed by two persistent challenges across both the planning and execution phases. The first is the inherent variability in maintenance activity completion times. Activity durations depend on component health state, crew availability, and field findings that deviate from the plan, yet current practice sizes contingency buffers through engineering judgment rather than data-driven uncertainty quantification. The second is unplanned activities emerging during outage execution, where a situation found during a maintenance, testing, or surveillance activity (with potential consequences on outage execution) requires to be addressed. This paper presents a cross-phase analytical framework that addresses both of these challenges using a single architecture built on historical condition report records, work order histories, and outage schedule data as primary inputs. In the planning phase, a proactive pipeline processes the multi-cycle condition report and work order history through a sequence of analytical stages that construct a component knowledge graph, assess component health-state trajectories by analyzing degradation trends across successive outage cycles, and compute a composite risk score combining degradation frequency, prior emergent work precedent, and component criticality. Components exhibiting escalating health deterioration are flagged even when no prior unplanned maintenance record exists, enabling the detection of first-failure components that would otherwise receive a zero risk score under conventional evidence-based methods. In the execution phase, a reactive pipeline accepts a condition report, and it produces a structured recommendation (proceed, defer, escalate, or monitor) supported by a duration estimate derived from the semantic retrieval of analogous historical work orders at the median and 80th-percentile levels via relevance-weighted empirical quantiles, alongside an analog count as a calibrated confidence signal.

IndusDiff: A Latent Diffusion Framework for Industrial Audio Generation for Data-Scarce PHM

Mon, 28 Sep 2026 00:00:00 +0000

Prognostics and Health Management (PHM) systems increasingly rely on data-driven models for fault diagnosis and anomaly detection, yet their effectiveness is fundamentally constrained by the scarcity and imbalance of labeled fault condition data. This challenge is particularly acute in industrial audio monitoring, where failure events are rare, hazardous to induce, and highly variable across operating conditions. While recent advances in generative modeling offer a promising pathway for data augmentation, existing audio generation methods either incur high computational cost in waveform-domain diffusion models or fail to preserve phase-sensitive characteristics critical for capturing transient fault signatures.

To address these challenges, this paper proposes IndusDiff, a phase-aware latent diffusion framework for high-fidelity industrial audio generation tailored for PHM applications. The proposed approach combines a phase-aware convolutional autoencoder with a latent diffusion model to enable efficient and physically meaningful audio synthesis. The autoencoder is trained using a multi-resolution spectral loss that jointly enforces waveform fidelity, spectral consistency, and temporal phase coherence, thereby preserving diagnostically relevant features such as transients, harmonics, and high-frequency components in the latent representation. Diffusion is then performed in this compressed latent space, significantly reducing computational complexity while maintaining generation quality.

Unlike conventional latent diffusion methods that primarily focus on perceptual realism, the proposed framework explicitly addresses the unique characteristics of industrial audio signals, including non-stationarity, multi-scale temporal dynamics, and phase-sensitive fault signatures. By preserving both magnitude and phase information, the model generates synthetic audio that is not only perceptually realistic but also structurally consistent with real machine signals, making it suitable for downstream PHM tasks.

The framework is evaluated on two complementary datasets: a publicly available industrial motor dataset and an in-house dataset capturing multi-stage assembly operations. Generation quality is assessed using the Frechet Audio Distance (FAD) as the primary quantitative metric, along with waveform and spectrogram analyses for qualitative validation. Experimental results demonstrate that IndusDiff produces statistically consistent, high-fidelity industrial audio while achieving significant improvements in computational efficiency, generating 48 kHz audio samples lasting several seconds in just seconds on modern GPU hardware.

DiFG-Net++: Physics-Informed Functional Learning for Self-Supervised and Domain-Adaptive Fault Diagnosis

David He, Miao He, Muhammad Areeb — Mon, 28 Sep 2026 00:00:00 +0000

Cross-speed fault diagnosis remains a significant challenge in data-driven prognostics and health management (PHM) because vibration signatures and fault characteristic frequencies vary systematically with rotational speed. The problem becomes particularly difficult in open-set scenarios where the target operating speed lies outside the range observed during training and fault data at the target speed are unavailable. Although recent advances in self-supervised learning (SSL), particularly Prediction of Functionals from Masked Latents (PFML), have demonstrated promising capabilities for learning fault-sensitive representations from unlabeled data, existing PFML-based frameworks do not explicitly address speed-induced distribution shifts. In particular, the recently developed Domain-Informed Functional Guidance Network (DiFG-Net) relies primarily on handcrafted functionals and lacks mechanisms for enforcing speed-invariant representation learning.

To address these limitations, this paper proposes DiFG Net++, a physics-informed functional learning framework specifically designed for open-set cross-speed fault diagnosis. Building upon the DiFG-Net architecture, DiFG Net++ integrates three complementary speed-adaptation strategies: (1) speed-extrapolative augmentation, which generates
physically meaningful intermediate and extrapolated speed conditions through speed-guided time scale transformations; (2) speed-invariant latent regularization, which encourages latent representations associated with the same machine health condition to remain consistent across operating speeds; and (3) healthy-state target-speed anchoring, which leverages readily available healthy vibration data at the target operating speed to facilitate adaptation without requiring target-speed fault samples. In addition, order-domain functionals are incorporated as physics-informed self-supervised targets to promote learning of speed-robust fault characteristics.

The effectiveness of the proposed framework is evaluated on a plastic-bearing fault dataset collected at four rotational speeds and across multiple bearing health conditions. Experimental results demonstrate that the proposed speed adaptation mechanisms substantially improve open-set cross-speed fault diagnosis performance. Compared with the baseline DiFG-Net, which achieved an overall validation accuracy of 46.68%, the proposed DiFG-Net++ achieved an overall accuracy of 95.85%, representing a relative improvement of more than 105%. These results demonstrate that integrating physics-informed functional learning with speed-aware representation learning provides an effective and practical solution for robust cross-speed fault diagnosis under previously unseen operating conditions.

Demonstration-Based Visual Anomaly Detection in Manufacturing Factory Using Vision-Language Models

Huimin Zhuge, Song Wang, Huijuan Shao, Brian Chien, Dipanjan Ghosh — Mon, 28 Sep 2026 00:00:00 +0000

Flexible anomaly detection is important for identifying process deviations in manufacturing, yet conventional approaches typically depend on hand-coded inspection rules, annotated datasets, task-specific model training, or additional sensing infrastructure, all of which are costly to configure and difficult to adapt. This work investigates demonstration based visual anomaly detection using vision-language models (VLMs), in which monitoring behavior is specified by example rather than by explicit programming. Instead of enumerating normal operating rules, the model is shown images and videos of nominal operation and infers the expected visual patterns directly from these demonstrations. During monitoring, new observations are assessed against the demonstrated patterns to detect potential abnormal states and to generate interpretable warnings and corrective suggestions. We formalize the detection problem mathematically and evaluate the approach on three representative fault types: sorting faults, where a color category is missing or misplaced; storage faults, where caps violate the expected bottom-to-top filling pattern or appear in an incorrect color region; and conveyor faults, where caps overlap, become stuck, move irregularly, or lose consistent spacing. These scenarios are realized on a small manufacturing cell equipped with pickup arms, a conveyor belt, sorting and storage units, and fixed monitoring cameras, using colored cylinder caps as representative workpieces. Results indicate that visual demonstrations combined with reasoning can substantially reduce the effort required to configure perception, anomaly interpretation, and decision support in manufacturing workflows.

Physics-Informed Condition Monitoring of SiC Power Modules

Mon, 28 Sep 2026 00:00:00 +0000

Silicon carbide (SiC) power modules are increasingly deployed in automotive traction inverters, where reliable condition monitoring is essential to prevent in-service failures and reduce maintenance costs. Despite extensive qualification procedures standardized under AQG 324, no consolidated approach exists for in-field health state estimation of SiC devices. Existing methods range from physics-of-failure lifetime models, which lack real-time applicability, to purely data-driven architectures that require large labeled datasets and offer limited generalization, to physics-informed machine learning frameworks that, while promising, remain computationally demanding for embedded deployment.

This work addresses condition monitoring of SiC MOSFET power modules assembled with sintered packaging technology, which prevents solder degradation and thereby produces aging behavior qualitatively distinct from previously studied devices. In solder-based modules, the forward voltage drop $V_{DS}$ follows smooth quasi-exponential trajectories driven by progressive solder delamination. With sintered packaging this mechanism is suppressed, and $V_{DS}$ instead exhibits multi-regime degradation profiles; modules are additionally subject to wirebond liftoff events that introduce abrupt, non-monotonic perturbations directly onto $V_{DS}$, posing new challenges for condition monitoring approaches.

To address these challenges, we propose a condition monitoring framework integrating three complementary elements. First, physics-informed feature engineering replaces raw sensor signals with physically grounded inputs, including cumulative damage indicators derived from junction temperature swing and mean junction temperature, combined with a Miner rule accumulator, which encode degradation history in a form directly interpretable by the model. Second, a monotonicity constraint enforced via gradient penalty regularization embeds the expected degradation direction as a physics-guided prior, improving generalization and cross-validation stability. Third, the model output is parameterized as a heavy-tailed distribution rather than a point estimate, providing calibrated uncertainty quantification inherently robust to the out-of-distribution variance introduced by liftoff events.

The framework is evaluated on an industrial power cycling dataset provided by Infineon Technologies, acquired under multiple operating conditions. Multiple neural network architectures of varying complexity and sequential context are compared within a rigorous cross-validation protocol. The full physics-informed feature set combined with gradient penalty regularization consistently outperforms purely data-driven baselines, reducing mean absolute error by approximately 70\% and maintaining stable performance across all cross-validation folds, including conditions where baseline models degrade significantly. These results confirm that integrating physical prior knowledge at the feature, constraint, and output distribution levels is both necessary and sufficient to handle the complexity of SiC power module degradation, while remaining lightweight enough for practical embedded deployment.

Uncertainty-Aware Model-Constrained Neural Digital Twins for Real-Time Forecast, Calibration, and Control of Aeroelastic Airfoils

Mon, 28 Sep 2026 00:00:00 +0000

Current structural health monitoring (SHM) in aeroelastic systems struggles with recovering unobservable states from limited sensor data, lacks quantified prediction confidence, and exhibits brittleness under off-nominal conditions. To address these operational gaps, this paper presents a physics aware neural digital twin (DT) designed for real-time, adaptive monitoring of aerospace structures. The proposed DT framework combines two neural networks for stabilizing control and forward prediction while using Bayesian filtering for nonlinear state estimation and uncertainty quantification. We train the controller using Proximal Policy Optimization (PPO) [1], an actor-critic method, and the forward model using the mcTangent loss [2]. We use Bayesian particle filter ing to infer the state and provide uncertainty quantification for assurance. The DT then provides the control input to the physical twin using the mean state estimate. Using only pitch angle measurements from a nonlinear airfoil model, the DT recovers full system state, stabilizes the structure rapidly, and provides uncertainty quantification (UQ) with all predictions. Finally, we compare the results to a more traditional approach of ensemble Kalman filter (EnKF) and model predictive control (MPC) and the base ODE model rather than the surrogate, which preforms worse due to worse early estimation of the hidden state. We also discuss the architecture’s transferability to broader digital twin applications across aerospace platforms.

A Sequential Bayesian Semi Markov Framework for Mission Based Prognostics

Mariana Salinas-Camus, Nick Eleftheroglou — Mon, 28 Sep 2026 00:00:00 +0000

Prognostics focuses on predicting the Remaining Useful Life (RUL) of engineering systems, which is a key enabler for informed decision-making in Predictive Maintenance (PdM) and operational planning. While substantial research has addressed maintenance strategies, the impact of operational decisions on system degradation and RUL remains relatively underexplored. In mission-based operations, systems may be subjected to different operating profiles, each influencing degradation dynamics. Understanding how operational choices affect RUL is therefore crucial for reliable mission planning.

This paper proposes a sequential Bayesian semi-Markov (SBSM) framework for prognostics that explicitly accounts for operational variability. System degradation is modelled using a hidden semi-Markov model (HSMM), a data-driven and unsupervised approach. Inference is performed via a particle filter (PF), enabling online estimation of hidden degradation states and RUL prediction under varying profiles.

The proposed framework is validated using simulated degradation data. During training, multiple operational profiles are available, with each run‑to‑failure trajectory conducted under a single, fixed operating profile. In contrast, the testing phase considers degradation histories in which the operating profile changes during system operation. Results demonstrate that the proposed SBSM approach can effectively track degradation and provide meaningful RUL estimates under varying operational profiles, highlighting its potential for mission‑based prognostics.

An End-to-End PHM Framework for Industrial Compressor Gearbox Fault Detection and RUL Prediction

Changik Cho, Eric Bechhoefer — Mon, 28 Sep 2026 00:00:00 +0000

Gearbox health monitoring in industrial machinery is essential for maintaining production efficiency, reducing maintenance costs, and ensuring reliable operation. However, the validation of complete prognostic frameworks on real-world industrial gearboxes remains limited compared to studies conducted on controlled laboratory test rigs. This study presents a full end-to-end Prognostics and Health Management (PHM) framework validated on a run-to-failure dataset collected from an operational industrial spur gearbox used in compressor machinery. Vibration data is acquired using a radially mounted accelerometer while shaft speed is derived from a 1/rev tachometer signal. From the processed signals, condition indicators are extracted across multiple vibration analysis domains, including residual signal analysis, amplitude and frequency demodulation, energy operator techniques, and sideband modulation analysis. From these algorithms, features sensitivity to gear fault modes such as tooth cracking, pitting, scuffing, and misalignment are calculated. These condition indicators are fused into a single scalar health index which can determine when maintenance should be performed. A Remaining Useful Life (RUL) is estimated using a high cycle fatigue model, allowing stable real-time prediction without requiring detailed material-specific parameters. The RUL provides the operator with actional information as to when best to schedule maintenance. Results demonstrate that the proposed framework enables early fault detection and provides consistent, physically interpretable RUL predictions, supporting condition-based maintenance decisions and improving operational planning in industrial gearbox systems.

Planning or Learning: Reliability and Cost in Multi-Asset Maintenance

Xian Yeow Lee, Chandrasekar Venkatraman, Ahmed Farahat — Mon, 28 Sep 2026 00:00:00 +0000

Industrial maintenance systems increasingly involve multiple interacting assets and shared resources, making it challenging to balance reliability and operational cost using a single decision framework. While recent work has focused on reinforcement learning (RL) for maintenance scheduling, direct comparisons with planning approaches under identical settings remain limited. In this work, we empirically compare planning and RL for multi-asset bearing maintenance using run-to-failure data. We examine how these methods behave when balancing preventive maintenance against tolerable failures across a range of failure penalty scenarios. We observed a consistent behavioral difference driven by objective formulation. Planning enforces reliability as a hard constraint and produces zero-failure policies whose total cost is largely insensitive to the magnitude of failure penalties. RL agents optimize expected cost and often trade off preventive maintenance against occasional failures as penalties vary, resulting in lower costs under low-penalty regimes but persistent nonzero failures even when penalties are high. We also investigate lightweight constraint mechanisms, including reward shaping and action masking, to encourage RL’s reliability. From a practical perspective, planning may be more suitable when strict reliability is required and deployment horizons are short, whereas RL may provide cost-efficient policie when limited failures are acceptable and long-run operational efficiency is prioritized. Overall, this study clarifies the tradeoffs between reliability and cost in multi-asset maintenance, highlights the challenges of enforcing zero-failure behavior in RL, and suggests that planning and RL are complementary approaches whose applicability depends on operational objectives. Beyond these findings, the controlled benchmark protocol itself that unifies environment, cost model, and evaluation across paradigms, offers a reusable template for comparing decision-making approaches in other maintenance settings.

Model-Based Online Detection, Quantification and Localization of Internal Short Circuits in Lithium-Ion Pouch Cells via an Unscented Kalman Filter

Lorenzo Branca, Yiqi Jia, Danyang Wang, Marco Giglio, Francesco Cadini — Mon, 28 Sep 2026 00:00:00 +0000

Internal short circuits (ISCs) are a leading precursor of thermal runaway in lithium-ion batteries, yet they are difficult to observe directly because their early electrical signature is small and ambiguous and their thermal signature is spatially distributed. This paper presents a model-based observer that simultaneously detects an ISC, quantifies its severity, and localizes it on the surface of a pouch cell using only the applied current, the terminal voltage, and a small array of surface temperature sensors. A lumped electro-thermal model couples a third-order equivalent-circuit model (ECM) of the cell to a two-dimensional lumped-parameter thermal network (LPTN); the ISC is parameterized by a shunt conductance G and a planar location (x, y), and is the source of an additional ohmic heat term injected locally into the thermal field. These three fault parameters are appended to the dynamic state and estimated online with a single joint unscented Kalman filter (UKF). A location-observability gate suppresses spurious position updates until the measured spatial thermal contrast is informative, and the detection instant is defined as the time at which the filter covariance converges to a stable value. The method is evaluated on a design of experiments (DoE) of subcritical shorts generated by a high-fidelity electrochemical–thermal plant, spanning three severities and nine locations. For strong and moderate shorts the observer recovers severity to within about 7% and location to within about 2 mm, converging in roughly 110 s; for the weakest short the severity is quantifiable only to order of magnitude and the thermal contrast is insufficient to localize, defining the detectability floor of the sensing arrangement.

Fault Detection and Remaining Useful Life Estimation of a Scuffed Gear

Eric Bechhoefer, Changik Cho, Lotfi Saidi — Mon, 28 Sep 2026 00:00:00 +0000

In helicopter drivetrains, vibration-based gear fault detection is critical to safety, reliability, and operational availability. Although there are numerous gear fault detection algorithms have been proposed, there is comparatively little work on gear wear.

This study presents a quantitative comparison of time synchronous average (TSA) based gear fault detection algorithms for identifying scuffing and wear in a turboshaft engine high-speed turbine shaft pinion. A large field dataset from a degraded engine is compared with data from a nominal replacement component. Condition indicator (CI) responses are evaluated using statistical separability to determine each algorithm’s ability to distinguish the damaged gear from the nominal baseline.

The comparison includes residual and difference-signal analysis, energy-operator methods and variants, narrowband analysis with bandwidth sensitivity evaluation, amplitude and frequency modulation analysis, and standard gear fault condition indicators. The statistical significance and robustness of each method are quantified.

The best-performing indicators are then used to construct a component health index (HI), which is applied to assess gear condition and demonstrate remaining useful life (RUL) estimation in a maintenance-relevant framework.

From Interoperability to Explainability: Semantic Data Models for XAI in Industrial Prognostics and Health Management

Anand Todkar, Mrinmoy Sarkar, Jitendra Solanki, Ziran Min — Mon, 28 Sep 2026 00:00:00 +0000

Semantic data models play a key role in enabling interoperability in industrial plants, and their importance has grown further with the rapid adoption of industrial artificial intelligence (AI). As AI applications become increasingly prevalent, most of them rely on data-driven approaches and exhibit a predominantly black-box nature, making their output difficult to interpret and explain. To address this limitation, this work investigates the integration of semantic data models with AI applications to enhance the interpretability of their results.

In this study, the OPC UA ISA-95 companion specification is utilized as the semantic data model, while a random forest–based AI model is applied for predictive maintenance in a manufacturing shop floor context. To support explainable AI (XAI), the results are evaluated in five dimensions: faithfulness, stability, sparsity, redundancy, and calibration. Furthermore, a comparative analysis is conducted under three scenarios: (i) using a flat tabular data set with feature engineering, (ii) incorporating the semantic model, and (iii) augmenting the semantic model with the flat tabular data set. Preliminary findings indicate that the integration of semantic data models enhances the explainability of black-box AI models.

Phantom Fires: Safe False-Alert Suppression via Multimodal Fusion of a Visible-Patch CNN and Solar Reflection Geometry for Thermal Fault Monitoring

Mon, 28 Sep 2026 00:00:00 +0000

Continuous outdoor thermal monitoring raises an analyst alert whenever an IR pixel exceeds a fixed apparent-temperature threshold. We report a 40-day deployment over 10 cameras (4,143.6 camera-hours) that produced 157 above-threshold events (≥ 150 °C): 122 (77.7 %) solar-glare false positives and 35 (22.3 %) confirmed true thermal events — a glare event rate of 0.0294 events per camera-hour (95 % CI [0.0157, 0.0449], bootstrap n = 1000). To suppress this false-alarm burden without compromising thermal-event recall, we construct a paired visible-patch dataset of 10,622 RGB patches and evaluate a ladder of five glare/noglare discriminators of increasing complexity, from a brightness threshold through a forward Sandia SGHAT geometric model, a logistic regression over 28 hand-crafted features, a small CNN (245,889 parameters), and a multimodal late fusion of the CNN and geometric scores. The fusion classifier reaches 98.6 % accuracy on a balanced 2,126-patch held-out eval set, a 3.3-percentage-point gain over the CNN alone and a 14.1-point gain over the geometric model; the learned weights (w_cnn = 7.34, w_geo = 4.50, bias = −1.93) confirm both streams carry independent signal, and adding the geometric stream cuts the CNN's false-positive count by 73 % while preserving recall. A separate system-level test on the 35 in-window true thermal events shows that both the CNN and the fusion classifier reach 99.58 % specificity (235 of 236 IR/VIS frame pairs preserved at the 150 °C production threshold). The current production deployment runs no glare suppression in place — every above-threshold IR event raises an analyst alert today — so the five-rung ladder and the multimodal fusion are research-grade results, not deployed artefacts. We propose a three-way routing policy (alert, fallback, suppress) that wraps the best rung and report the measured anchors that constrain its expected outcome on the deployment ledger.

Applied Structured Corrective Diagnosis of a Legacy Autonomous Mobile Robot:

Zen Hassoun, Hazem Eissa — Mon, 28 Sep 2026 00:00:00 +0000

Corrective diagnosis of legacy robotic assets is routinely performed in operating contexts where manufacturer documentation is incomplete, telemetry is unavailable, and the failure mode is non-progressive. These operations fall outside the typical scope of both data-driven and knowledge-based PHM techniques. This paper presents a structured framework for diagnosing legacy AMR startup failures under limited information conditions, together with a comparison of three diagnostic prioritisation strategies applied to the same practical case: (i) a qualitative power-tracing logic that combines Fault Tree Analysis (FTA), Failure Mode and Effects Analysis (FMEA), and a decision flowchart, which was the method applied to the asset; (ii) an analytical FMEA with Risk Priority Number (RPN) scoring; and (iii) an analytical decision cost ordering that ranks investigation steps by expected total cost. Methods (ii) and (iii) are then applied retrospectively to the same case using the operator’s experience gained during the practical investigation to assign scores and parameters. The case is a no-start fault on a legacy MiR100 autonomous mobile robot in a university laboratory. The practical investigation, following method (i), traced the fault to a blown fuse in the robot-side input path; replacement and a brief charging recovery restored normal operation, confirmed via the built-in hardware-health diagnostic. Applied retrospectively, method (ii) would have ordered the investigation by descending RPN, reaching the fuse on the seventh step, method (iii) would have ordered by the lowest per-step decision cost, reaching the fuse within minutes. We compare the three methods on time-to-confirmation and total cost, and found that methods (i) and (ii) reach the fault with comparable total time and cost despite different orderings, while method (iii) reaches it approximately an order of magnitude faster and cheaper in this case.

A Functional Analysis Approach for Prototyping a Prognostic Model

Darin Dunlap, Jonathan Lawrence, Brandon Osborn, Taylor Spoo, Kyle Turner — Mon, 28 Sep 2026 00:00:00 +0000

Large datasets in prognostics and health management (PHM) can often present challenges for degradation prediction models because of size and computational cost. In this work, a methodology is introduced that employs functional data analysis (FDA) to compress high dimensional sensor data using a B-spline basis. This method was able to reduce the dimensionality of the data while preserving the essential characteristics of the underlying trend. The reduced dimensional representation serves as input to a k-Nearest Neighbors (KNN) algorithm for remaining useful life (RUL) estimation of a jet engine nacelle fan. Experimental results demonstrate that the functional KNN (fKNN) framework can predict part failure within 50 flight hours, achieving sufficiently reliable and precise RUL forecasts on the test data set. The optimal hyperparameters as validated by the test set used 25 nearest neighbors, a prediction count of 3, and a prediction minimum threshold of 0.44. Therefore, the combined use of FDA for dimensionality reduction and KNN for RUL prediction offers a scalable solution for handling big data (~20 GB) PHM challenges without sacrificing predictive performance.

Enhancing PHM Functionality by Using Data Sharing Standards

Ravi Rajamani, Leon Gommans — Mon, 28 Sep 2026 00:00:00 +0000

Nearly everything related to Prognostics and Health Management (PHM) in aviation is dependent on data: its generation, its transmission, its storage and its access and use. In each of these areas, many standards exist related to the underlying technologies, but this is not the case when it comes to standards for data governance. In this brief paper, we address the gaps in the landscape on data governance, especially those related to data sharing. This will be critical when artificial intelligence and in particular machine learning becomes more widespread in developing and maintaining PHM algorithms. We describe how aerospace consortia may be able to help fill some of these gaps. The path forward in this highly connected world will only be possible if strict rules exist for all entities within the ecosystem to be able to share data and thereby achieve common benefits no single organization can achieve on its own. In particular, we describe the role of the Independent Data Consortium for Aviation (IDCA) that has been established to do just this. Some examples of areas within aviation where governance standards can be critical are parts data tracking, AOG (aircraft on ground) situations, diagnostics and prognostics for aircraft systems, supporting the use of artificial intelligence (AI) and machine learning (ML). We show how issues can be resolved faster and more efficiently if data governance standards exist that are agreed to by all stakeholders. Additionally, mechanisms for ensuring trustworthiness would be critical in allowing automated enforcement and accounting. Here we are reporting on the progress we are making in developing these standards.