In this paper a modification to a general-purpose machinery diagnostic/prognostic algorithm that can handle two or more simultaneously occurring failure processes is described. The method is based on a theory that views damage as occurring in a hierarchical dynamical system where slowly evolving, hidden failure processes are causing nonstationarity in a fast, directly observable system. The damage variable tracking is based on statistics calculated using data-based local linear models constructed in the reconstructed phase space of the fast system. These statistics are designed to measure a local change in the fast systems flow caused by the slow-time failure processes. The method is applied to a mathematical model of an experimental electromechanical system consisting of a beam vibrating in a potential field crated by two electromagnets. Two failure modes are introduced through discharging batteries supplying power to these electromagnets. Open circuit terminal voltage of these batteries is a two-dimensional damage variable. Using computer simulations, it is demonstrated both analytically and experimentally that the proposed method can accurately track both damage variables using only a displacement measurements from the vibrating beam. The accurate estimates of remaining time to failure for each battery are given well ahead of actual breakdowns.
This paper discusses some of the major results from the initial effort of the Predictive Failures and Advanced Diagnostics for Legacy Aircraft program. The primary goals of this AF Research Laboratory/Northrup Grumman project are to enable a prognostics capability for legacy avionics systems and to enhance their existing diagnostic performance. Major benefits of this program are enhanced aircraft availability ad reduced Operation and Support costs.
As the health situation of a system is only indirectly accessible, often conclusive explanations for observed abnormal behavior can not be given. In order to discriminate further between possible diagnoses, more information about system behavior is necessary. Testing techniques are especially useful in situations where it is not possible to probe additional process variables, such as in remote diagnosis applications. However as Scarl has pointed out, care must be taken as test vectors may induce new errors. He introduced the notion of so-called hazard condition constraints that should not be violated by the test input. In this paper, we apply the notion of safe test vector generation to the domain of dynamic systems. Dynamic systems are characterized by the fact that the current behavior does not depend on the current input only, but also on the history of the system. Therefore, safe testing for dynamic systems needs a technique akin to model-predictive control. That is, before one can say that a particular test vector will discriminate between two possible diagnoses, or that it will not violate a hazard condition, the behavior of the system has to be simulated over a number of time steps.
Proc. SPIE 4733, Integration of health management and support systems is key to achieving cost reduction and operational concept goals of the 2nd generation reusable launch vehicle, 0000 (16 July 2002); doi: 10.1117/12.475511
Our aerospace customers are demanding that we drastically reduce the cost of operating and supporting our products. Our space customer in particular is looking for the next generation of reusable launch vehicle systems to support more aircraft like operation. To achieve this goal requires more than an evolution in materials, processes and systems, what is required is a paradigm shift in the design of the launch vehicles and the processing systems that support the launch vehicles. This paper describes the Automated Informed Maintenance System (AIM) we are developing for NASA's Space Launch Initiative (SLI) Second Generation Reusable Launch Vehicle (RLV). Our system includes an Integrated Health Management (IHM) system for the launch vehicles and ground support systems, which features model based diagnostics and prognostics. Health Management data is used by our AIM decision support and process aids to automatically plan maintenance, generate work orders and schedule maintenance activities along with the resources required to execute these processes. Our system will automate the ground processing for a spaceport handling multiple RLVs executing multiple missions. To accomplish this task we are applying the latest web based distributed computing technologies and application development techniques.
The NASA Integrated Vehicle Health Management (IVHM) Technology Experiment for X-37 was intended to run IVHM software on board the X-37 spacecraft. The X-37 is an unpiloted vehicle designed to orbit the Earth for up to 21 days before landing on a runway. The objectives of the experiment were to demonstrate the benefits of in-flight IVHM to the operation of a Reusable Launch Vehicle, to advance the Technology Readiness Level of this IVHM technology within a flight environment, and to demonstrate that the IVHM software could operate on the Vehicle Management Computer. The scope of the experiment was to perform real-time fault detection and isolation for X-37's electrical power system and electro-mechanical actuators. The experiment used Livingstone, a software system that performs diagnosis using a qualitative, model-based reasoning approach that searches system-wide interactions to detect and isolate failures. Two of the challenges we faced were to make this research software more efficient so that it would fit within the limited computational resources that were available to us on the X-37 spacecraft, and to modify it so that it satisfied the X-37's software safety requirements. Although the experiment is currently unfunded, the development effort resulted in major improvements in Livingstone's efficiency and safety. This paper reviews some of the details of the modeling and integration efforts, and some of the lessons that were learned.
The operations of aircraft fleets typically result in large volumes of data collected during the execution of various operational and support processes. This paper reports on an Army-sponsored study conducted to research the applicability of data mining for processing such data. The study focused on three aspects: (1) understanding the aviation operations, maintenance environment, and data collection system; (2) investigating data analysis approaches with the purpose of identifying promising methods pertinent to aircraft health management; and (3) defining requirements for a tool to support the aviation maintenance planners and fleet managers. Results of preliminary analyses of two maintenance data and flight data sets are presented. An architecture for managing and mining aviation maintenance data and using results to update models used by diagnostic modules for fault isolation during maintenance activity is also presented.
The majority diagnostic research of rotating machinery consists in observation and analysis of vibration signals recorded during its operation. In complex machinery case interpretation of results of these signal analysis is difficult and there is no methods of their automatic interpretation. The paper deals with concept of application of system that is based on notation of results of analysis in form of dynamic scene. It makes it possible to consider signal features from different points of view. That approach lets us also to apply simple methods of image analysis that we use in change identification. Determined changes are the basis of concluding process that is performed at two stages.
In this paper, the Bayesian Data Reduction Algorithm is applied to a collection of medical diagnostic data sets found at the University of California at Irvine's Repository of Machine Learning databases. The algorithm works by finding the best performing quantization complexity of the feature vectors, and this makes it necessary to discretize all continuous valued features. Therefore, results are given by showing the quantization of the continuous valued features that yields best performance. Further, the Bayesian Data Reduction Algorithm is also compared to a conventional linear classifier, which does not discretize any feature values. In general, the Bayesian Data reduction Algorithm is shown to outperform the linear classifier by obtaining a lower probability of error, as averaged over all data sets.
Classifier performance evaluation is an important step in designing diagnostic systems. The purposes of performing classifier performance evaluation include: 1) to select the best classifiers from the several candidate classifiers, 2) to verify that the classifier designed meets the design requirement, and 3) to identify the need for improvements in the classifier components. In order to effectively evaluate classifier performance, a classifier performance measure needs to be defined that can be used to measure the goodness of the classifiers considered. This paper first argues that in fault diagnostic system design, commonly used performance measures, such as accuracy and ROC analysis are not always appropriate for performance evaluation. The paper then proposes using misclassification cost as a general performance measure that is suitable for binary as well as multi-class classifiers, and -most importantly- for classifiers with unequal cost consequence of the classes. The paper also provides strategies for estimating the cost matrix by taking advantage of fault criticality information obtained from FMECA. By evaluating the performance of different classifiers considered during the design process of an engine fault diagnostic system, this paper demonstrates that misclassification cost is an effective performance measure for evaluating the performance of multi-class classifiers with unequal cost consequence for different classes.
In this study, we propose a novel hybrid intelligent system (HIS) which provides a unified integration of numerical and linguistic knowledge representations. The proposed HIS is hierarchical integration of an incremental learning fuzzy neural network (ILFN) and a linguistic model, i.e., fuzzy expert system, optimized via the genetic algorithm. The ILFN is a self-organizing network with the capability of fast, one-pass, online, and incremental learning. The linguistic model is constructed based on knowledge embedded in the trained ILFN or provided by the domain expert. The knowledge captured from the low-level ILFN can be mapped to the higher-level linguistic model and vice versa. The GA is applied to optimize the linguistic model to maintain high accuracy, comprehensibility, completeness, compactness, and consistency. After the system being completely constructed, it can incrementally learn new information in both numerical and linguistic forms. To evaluate the system's performance, the well-known benchmark Wisconsin breast cancer data set was studied for an application to medical diagnosis. The simulation results have shown that the prosed HIS perform better than the individual standalone systems. The comparison results show that the linguistic rules extracted are competitive with or even superior to some well-known methods.
The paper deals with methods of interpretation of features of vibration signals, which make it possibly to detect changes of these features. Considered signals are mainly recorded during varying conditions of rotating machinery. Signal features are coded in matrices that are similar to image data. Matrix is considered as dynamic scene and signal features are treated as that scene objects. These methods make it possible to identify variability of features that are strictly related to changes of technical state of the machinery. Scene analysis may be carried out with the use of methods of image analysis. Examples of these methods are lines detection or syntactic description. There is also an example of application of technical state observers.
A physics-based approach for diagnostics and prognostics using integrated observers and life models is presented. Observers are filters based on physical models of machine- fault combinations and use measured machine signatures to identify and characterize the state of a machine. Observers are adaptively deployed as a machine wears and can be coupled with one another to handle interacting conditions and faults. The scheme is detailed using the fault of a cracked rotor shaft that interacts with gravity and imbalance. Observers for shaft cracking and imbalance are presented. The observers provide machine condition and fault strengths to life models used to determine remaining machine life. A life model based on the Forman crack growth law of linear elastic fracture mechanics is developed to determine the number of machine cycles remaining until catastrophic failure.
Traditional static maintenance scheduling based on lifetime data and replacement upon failure is adequate for typical power users. However, in the case of high reliability/availability-oriented industries (e.g., power systems for internet data centers have a desired availability of 0.99999 and, for semiconductor fabrication plants, have availability requirement of 0.9999999), this type of preventive maintenance scheduling is inadequate. A suitable approach in these situations is the adoption of condition-based predictive maintenance. Here the system condition is evaluated by processing the information gathered from the monitors placed at different points in the system, and maintenance is performed only when the failure/malfunction prognosis dictates. In the past, for power systems, voltages, currents, power, temperature and electromagnetic quantities had been monitored along with surface inspection and material quality tests at regular intervals. Diagnostic methods are already in place to indicate problems in industrial power systems by examining these monitored quantities. However, they lack the capability of looking into distant future. With the introduction of modern digital electronics-based smart monitors, the capability of logging power quality data at micro-second intervals, advanced signal processing tools for extracting features from collected data, and data mining techniques, a new horizon in maintenance scheduling has been unveiled. Trending techniques and techniques based on neural networks, when applied to the extracted features, enable us to predict the possible failures of individual equipment and subsystems well before they manifest. This paper considers the problem of evaluating the health indices of components of a power system by making use of the monitored power-quality data and classification techniques. Health index analysis distinguishes the healthy and risky components of the system. Results of these evaluations can be fed as inputs into a system-reliability/availability analysis tool. The reliability analysis enables analysts to decide on prioritization of the maintenance options subject to budget constraints.
Inference of the expected time-to-failure is made difficult by the need to track and predict the trajectories of real-valued system parameters over essentially unbounded domains, and by the need to identify a subset of these domains that refers to a state of unsafe operation. In a previous paper we proposed a novel technique whereby these problems are avoided: instead of physical system or sensor parameters, sensor-level test-failure probability vectors (bounded within the unit hypercube) are tracked; and via a close relationship with the TEAMS suite of modeling tools, the terminal states for all such vectors can be enumerated. In that paper a full-dimension Kalman filter and IMM (interacting multiple model) tracking solution was proposed, but results were preliminary. In this paper we continue, modify, and provide reasonably convincing results.
This paper presents a novel methodology for the diagnosis and prognosis of crucial gear faults, such as gear tooth fatigue cracking. Currently, an effective detection of tooth cracking can be achieved by using the autoregressive (AR) modeling approach, where the gear vibration signal is modeled by an AR model and gear tooth cracking is detected by identifying the sudden changes in the model's error signal. The model parameters can be estimated under the criteria of minimum power or maximum kurtosis of model errors. However, these model parameters possess no physical meaning about the monitored gear system. It is proposed that the AR model be replaced by a gear dynamics model (GDM) that contains physically meaningful parameters, such as mass, damping and stiffness. By identifying and tracking the changes in the parameters, it is possible to make diagnosis and prognosis of gear faults. For example, a reduction in mesh stiffness may indicate cracking of a gear tooth. Towards physical model-based prognosis, an adaptive (or optimization) strategy has been developed for approximating a gear signal using a simplified gear signal model. Preliminary results show that this strategy provides a feasible adaptive process for updating model parameters based on measured gear signal.
Graph-based systems are models wherein the nodes represent the components and the edges represent the fault propagation between the components. For critical systems, some components are equipped with smart sensors for on-board system health management. When an abnormal situation occurs, alarms will be triggered from these sensors. This paper considers the problem of identifying the set of potential failure sources from the set of ringing alarms in graph-based systems. However, the computational complexity of solving the optimal multiple fault diagnosis problem is super-exponential. Based on Lagrangian relaxation and subgradient optimization, we present a heuristic algorithm to find the most likely candidate fault set. A computationally cheaper heuristic algorithm - primal heuristic - has also been applied to the problem so that real-time multiple fault diagnosis in systems with several thousand failure sources becomes feasible in a fraction of a second.
This work aims to establish a nonlinear dynamics framework for diagnosis and prognosis in structural dynamic systems. The objective is to develop an analytically sound means for extracting features, which can be used to characterize damage, form modal-based input-output data in complex hybrid structures with heterogeneous materials and many components. Although systems like this are complex in nature, the premise of the work here is that damage initiates and evolves in the same phenomenological way regardless of the physical system according to nonlinear dynamic processes. That is, bifurcations occur in healthy systems as a result of damage. By projecting a priori the equations of motion of high-dimensional structural dynamic systems onto lower dimensional center, or so-called 'damage', manifolds, it is demonstrated that model reduction near bifurcations might be a useful way to identity certain features in the input- output data that are helpful in identifying damage. Normal forms describing local co-dimension one and two bifurcations are assumed to govern the initiation and evolution of damage in a low-order model. Real-world complications in damage prognosis involving spatial bifurcations, global bifurcation phenomena, and the sensitivity of damage to small changes in initial conditions are also briefly discussed.
We propose to use characteristic exponents of averaged instability for diagnostics of complex systems. It is shown that these exponents can be calculated from eigenvalues of the product matrix obtained from Jacobian-matrixes of the multi-step transformation. The effective scheme for calculating the product matrix in a finite time interval is developed, it allows to reduce calculation complexity due to exact factorization of the Jacobian-matrix in restricted time intervals. The detailed analytical analysis of the reconstruction process is implemented with respect to delay nonlinear equations for both Runge-Kutta and Euler approaches.
The paper deals with selected methods of knowledge acquisition for intelligent information systems that may be applied for aiding technical diagnostics of machinery and equipment. Two main kinds of knowledge are discussed, i.e. declarative and procedural knowledge. Classification of knowledge acquisition methods is presented with respect to main knowledge sources. Some methods of knowledge acquisition from domain experts and from databases are described. A process of declarative knowledge acquisition is discussed in detail. Examples of applications of the knowledge acquisition methods are shown. The paper concludes with new issues of knowledge acquisition methodology.