Gregory J. Kacprzynski

Advances in uncertainty representation and management for particle filtering applied to prognostics

Particle filters (PF) have been established as the de facto state of the art in failure prognosis. They combine advantages of the rigors of Bayesian estimation to nonlinear prediction while also providing uncertainty estimates with a given solution. Within the context of particle filters, this paper introduces several novel methods for uncertainty representations and uncertainty management. The prediction uncertainty is modeled via a rescaled Epanechnikov kernel and is assisted with resampling techniques and regularization algorithms. Uncertainty management is accomplished through parametric adjustments in a feedback correction loop of the state model and its noise distributions. The correction loop provides the mechanism to incorporate information that can improve solution accuracy and reduce uncertainty bounds. In addition, this approach results in reduction in computational burden. The scheme is illustrated with real vibration feature data from a fatigue-driven fault in a critical aircraft component.

Assessment of data and knowledge fusion strategies for prognostics and health management

Various data, feature and knowledge fusion strategies and architectures have been developed over the last several years for improving upon the accuracy, robustness and overall effectiveness of anomaly, diagnostic and prognostic technologies. Fusion of relevant sensor data, maintenance database information, and outputs from various diagnostic and prognostic technologies has proven effective in reducing false alarm rates, increasing confidence levels in early fault detection, and predicting time to failure or degraded condition requiring maintenance action. The data fusion strategies discussed are principally probabilistic in nature and are used to aid in directly identifying confidence bounds associated with specific component fault identifications and predictions. Dempster-Shafer fusion, Bayesian inference, fuzzy-logic inference, neural network fusion and simple weighting/voting are the algorithmic approaches that are discussed. Data fusion architectures such as centralized fusion, autonomous fusion, and hybrid fusion are described in terms of their applicability to fault diagnosis and prognosis. The final goal is to find the optimal combination of measured system data, data fusion algorithms, and associated architectures for obtaining the highest overall prediction/detection confidence levels associated with a specific application. To achieve this goal, a set of metrics has been developed for gauging the performance and effectiveness of a fusion strategy. Specifically, this paper demonstrates how various metrics are used for assessing individual and fused vibration-based diagnostic algorithms. Evaluation of the diagnostic strategies was performed using gearbox seeded-fault and accelerated failure data.

Enhancement of physics-of-failure prognostic models with system level features

To truly optimize the deployment of DoD assets, there exists a fundamental need for predictive tools that can reliably estimate the current and reasonably predict the future capacity of complex systems. Prognosis, as in all true predictions, has inherent uncertainty, which has been treated through probabilistic modeling approaches. The novelty in the current prognostic tool development is that predictions are made through the fusion of stochastic physics-of-failure models, relevant system or component level health monitoring data and various inspection results. Regardless of the fidelity of a prognostic model or the quantity and quality of the seeded fault or run-to-failure data, these models should be adaptable based on system health features such as vibration, temperature, and oil analysis. The inherent uncertainties and variability in material capacity and localized environmental conditions, as well as the realization that complex physics-of-failure understanding will always possess some uncertainty, all contribute to the stochastic nature of prognostic modeling. However, accuracy can be improved by creating a prognostic architecture instilled with the ability to account for unexpected damage events, fuse with diagnostic results, and statistically calibrate predictions based on inspection information and real-time system level features. In this paper, the aforementioned process is discussed and implemented first on controlled failures of single spur gear teeth and then on a helical gear contained within a drivetrain system. The stochastic, physics-of-failure models developed are validated with transitional run-to-failure data developed at Penn State ARL. Future work involves applying the advanced prognostics process to helicopter gearboxes.

Methodologies for uncertainty management in prognostics

Effective uncertainty management processes are essential elements in the design of prognostic modules if they to be viable for Integrated Vehicle Health Management (IVHM) systems. Modeling uncertainty, measurement and estimation uncertainties, future load uncertainty, among other factors, all potentially contribute to prognostic uncertainty. This paper analyzes the source of uncertainties in typical IVHM systems and presents a rigorous set of algorithms for uncertainty management that are generic and capable of addressing a variety of uncertainty sources. Specifically, model parameter uncertainty is addressed by a Bayesian-based updating scheme with two variants. One approach utilizes an inner-outer loop Monte Carlo simulation scheme with hyper-parameter adaptation and is intended for off-line applications, while the other particle filtering-based approach can be implemented on-line in real-time. Modeling uncertainty (or model structure uncertainty) is addressed by a Bayesian model selection/fusion method. Effective approaches for handling diagnostic uncertainty and the aggregation of component level uncertainty to system level are also addressed. Select results for the application of particular algorithms are presented.

An Overview of Selected Prognostic Technologies With Application to Engine Health Management

The DoD has various vehicle platforms powered by high performance gas turbine engines that would benefit greatly from predictive health management technologies that can detect, isolate and assess remaining useful life of critical line replaceable units (LRUs) or subsystems. In order to meet these needs for next generation engines, dedicated prognostic algorithms must be developed that are capable of operating in an autonomous and real-time engine health management system software architecture that is distributed in nature. This envisioned prognostic and health management system should allow engine-level reasoners to have visibility and insight into the results of local diagnostic and prognostic technologies implemented down at the LRU and subsystem levels. To accomplish this effectively requires an integrated suite of prognostic technologies that can be applied to critical engine systems and can capture fault/failure mode propagation and interactions that occur in these systems, all the way up through the engine and eventually vehicle level. In the paper, the authors will present a generic set of selected prognostic algorithm approaches that can be applied to gas turbine engines, as well as provide an overview of the required reasoning architecture needed to integrate the prognostic information across the engine.

An integrated approach to helicopter planetary gear fault diagnosis and failure prognosis

This paper introduces the design of an integrated framework for on-board fault diagnosis and failure prognosis of a helicopter transmission component, and describes briefly its main modules. It suggests means to (1) validate statistically and pre-process sensor data (vibration), (2) integrate model-based diagnosis and prognosis, (3) extract useful features or condition indicators from data de-noised by blind deconvolution, and (4) combine Bayesian estimation algorithms and measurements to detect and identify the fault and predict remaining useful life with specified confidence and minimum false alarms.

Improved diagnostic and prognostic assessments using health management information fusion

Various data, feature and knowledge fusion strategies and architectures have been developed over the last several years for improving upon the accuracy, robustness and overall effectiveness of anomaly, diagnostic and prognostic technologies. Fusion of relevant sensor data, maintenance database information, and outputs from various diagnostic and prognostic technologies has proven effective in reducing false alarm rates, increasing confidence levels in early fault detection, and predicting time to failure or degraded condition requiring maintenance action. The data fusion strategies discussed in this paper are principally probabilistic or AI-based in nature and are used to aid in directly identifying confidence bounds associated with specific component fault identifications and predictions. Dempster-Shafer fusion, Bayesian inference, fuzzy-logic inference, neural network fusion and simple weighting/voting are the algorithmic approaches that are discussed in this paper. Data fusion architectures such as centralized fusion, autonomous fusion, and hybrid fusion are described in terms of their applicability to fault diagnosis and prognosis. The final goal is to find the optimal combination of measured system data, data fusion algorithms, and associated architectures for obtaining the highest overall prediction/detection confidence levels associated with a specific application. A gas turbine engine test cell sensor validation example is provided in the paper that is specifically related data fusion approaches for test cell sensor validation using Dempster-Shafer fusion.

A Prognostic Modeling Approach for Predicting Recurring Maintenance for Shipboard Propulsion Systems

Accurate prognostic models and associated algorithms that are capable of predicting future component failure rates or performance degradation rates for shipboard propulsion systems are critical for optimizing the timing of recurring maintenance actions. As part of the Naval maintenance philosophy on Condition Based Maintenance (CBM), prognostic algorithms are being developed for gas turbine applications that utilize state-of-the-art probabilistic modeling and analysis technologies. Naval Surface Warfare Center, Carderock Division (NSWCCD) Code 9334 has continued interest in investigating methods for implementing CBM algorithms to modify gas turbine preventative maintenance in such areas as internal crank wash, fuel nozzles and lube oil filter replacement. This paper will discuss a prognostic modeling approach developed for the LM2500 and Allison 501-K17 gas turbines based on the combination of probabilistic analysis and fouling test results obtained from NSWCCD in Philadelphia. In this application, the prognostic module is used to assess and predict compressor performance degradation rates due to salt deposit ingestion. From this information, the optimum time for on-line waterwashing or crank washing from a cost/benefit standpoint is determined.Copyright

Prognostics-enhanced Automated Contingency Management for advanced autonomous systems

This paper introduces a novel prognostics-enhanced automated contingency management (or ACM+P) paradigm based on both current health state (diagnosis) and future health state estimates (prognosis) for advanced autonomous systems. Including prognostics in ACM system allows not only fault accommodation, but also fault mitigation via proper control actions based on short term prognosis, and moreover, the establishment of a long term operational plan that optimizes the utility of the entire system based on long term prognostics. Technical challenges are identified and addressed by a hierarchical ACM+P architecture that allows fault accommodation and mitigation at various levels in the system ranging from component level control reconfiguration, system level control reconfiguration, to high level mission re-planning and resource redistribution. The ACM+P paradigm was developed and evaluated in a high fidelity unmanned aerial vehicle (UAV) simulation environment with flight-proven baseline flight controller and simulated diagnostics and prognostics of flight control actuators. Simulation results are presented. The ACM+P concept, architecture and the generic methodologies presented in this paper are applicable to many advanced autonomous systems such as deep space probes, unmanned autonomous vehicles, and military and commercial aircrafts.

Extending FMECA-health management design optimization for aerospace applications

Health management is a philosophy that merges component and system level health monitoring, consisting of anomaly detection, diagnostic and prognostic technologies, with the operations and maintenance arenas. The concepts of health management, in particular health monitoring system design, have not traditionally been an integral aspect of the overall system design process. This may be partly due to the fact that detailed cost/benefit analysis of health management system configurations cannot be fully realized at this stage. This paper presents an approach that extends the traditional Failure Mode, Effects and Criticality Analysis (FMECA) to create a virtual environment in which Health Monitoring architectures can be evaluated from a cost/benefit standpoint. This health monitoring system design strategy allows for the inclusion of sensors and diagnostic/prognostic technologies to be generated from traditional FMECA information. This approach also introduces an environment for enhanced realization of component design requirements and the anomaly, diagnostic, and prognostic technologies themselves.