This paper presents a distributed Bayesian fault diagnosis scheme for physical systems. Our diagnoser design is based on a procedure for factoring the global system bond graph (BG) into a set of structurally observable bond graph fac- tors (BG-Fs). Each BG-F is systematically translated into a corresponding DBN Factor (DBN-F), which is then used in its corresponding local diagnoser for quantitative fault detec- tion, isolation, and identification. By construction, the ran- dom variables in each DBN-F are conditionally independent of the random variables in all other DBN-Fs, given a subset of communicated measurements considered as system inputs. Each DBN-F and BG-F pair is used to derive a local diag- noser that generates globally correct diagnosis results by lo- cal analysis. Together, the local diagnosers diagnose all single faults of interest in the system. We demonstrate on an electri- cal system how our distributed diagnosis scheme is compu- tationally more efficient than its centralized counterpart, but without compromising the accuracy of the diagnosis results.

One of NASA’s key mission requirements is robust state estimation. Sensing, using a wide range of sensors and sensor fusion approaches, plays a central role in robust state estimation, and there is a need to diagnose sensor failure as well as component failure. Sensor validation techniques address this problem: given a vector of sensor readings, decide whether sensors have failed, therefore producing bad data. We take in this paper a probabilistic approach, using Bayesian networks, to diagnosis and sensor validation, and investigate several relevant but slightly different Bayesian network queries. We emphasize that on-board inference can be performed on a compiled model, giving fast and predictable execution times. Our results are illustrated using an electrical power system, and we show that a Bayesian network with over 400 nodes can be compiled into an arithmetic circuit that can correctly answer queries in less than 500 microseconds on average. Reference: O. J. Mengshoel, A. Darwiche, and S. Uckun, "Sensor Validation using Bayesian Networks." In Proc. of the 9th International Symposium on Artificial Intelligence, Robotics, and Automation in Space (iSAIRAS-08), Los Angeles, CA, 2008. BibTex Reference: @inproceedings{mengshoel08sensor, author = {Mengshoel, O. J. and Darwiche, A. and Uckun, S.}, title = {Sensor Validation using {Bayesian} Networks}, booktitle = {Proceedings of the 9th International Symposium on Artificial Intelligence, Robotics, and Automation in Space (iSAIRAS-08)}, year = {2008} }

The purpose of this project is to improve the accuracy of statistical software by providing reference datasets with certified computational results that enable the objective evaluation of statistical software. Currently datasets and certified values are provided for assessing the accuracy of software for univariate statistics, linear regression, nonlinear regression, and analysis of variance. The collection includes both generated and 'real-world' data of varying levels of difficulty. Generated datasets are designed to challenge specific computations. These include the classic Wampler datasets for testing linear regression algorithms and the Simon & Lesage datasets for testing analysis of variance algorithms. Real-world data include challenging datasets such as the Longley data for linear regression, and more benign datasets such as the Daniel & Wood data for nonlinear regression. Certified values are 'best-available' solutions. The certification procedure is described in the web pages for each statistical method. Datasets are ordered by level of difficulty (lower, average, and higher). Strictly speaking the level of difficulty of a dataset depends on the algorithm. These levels are merely provided as rough guidance for the user. Producing correct results on all datasets of higher difficulty does not imply that your software will pass all datasets of average or even lower difficulty. Similarly, producing correct results for all datasets in this collection does not imply that your software will do the same for your particular dataset. It will, however, provide some degree of assurance, in the sense that your package provides correct results for datasets known to yield incorrect results for some software. The Statistical Reference Datasets is also supported by the Standard Reference Data Program.

Python module "optbayesexpt" uses optimal Bayesian experimental design methods to control measurement settings in order to efficiently determine model parameters. Given a parametric model - analogous to a fitting function - Bayesian inference uses each measurement "data point" to refine model parameters. Using this information, the software suggests measurement settings that are likely to efficiently reduce uncertainties. A TCP socket interface allows the software to be used from experimental control software written in other programming languages. Code is developed in python and shared via GitHub's USNISTGOV organization.

Python module 'optbayesexpt' uses optimal Bayesian experimental design methods to control measurement settings in order to efficiently determine model parameters. Given an parametric model - analogous to a fitting function - Bayesian inference uses each measurement 'data point' to refine model parameters. Using this information, the software suggests measurement settings that are likely to efficiently reduce uncertainties. A TCP socket interface allows the software to be used from experimental control software written in other programming languages. Code is developed in Python, and shared via GitHub's USNISTGOV organization.

Python module 'optbayesexpt' uses optimal Bayesian experimental design methods to control measurement settings in order to efficiently determine model parameters. Given an parametric model - analogous to a fitting function - Bayesian inference uses each measurement 'data point' to refine model parameters. Using this information, the software suggests measurement settings that are likely to efficiently reduce uncertainties. A TCP socket interface allows the software to be used from experimental control software written in other programming languages. Code is developed in Python, and shared via GitHub's USNISTGOV organization.

As fault diagnosis and prognosis systems in aerospace applications become more capable, the ability to utilize information supplied by them becomes increasingly important. While certain types of vehicle health data can be effectively processed and acted upon by crew or support personnel, others, due to their complexity or time constraints, require either automated or semi-automated reasoning. Prognostics-enabled Decision Making (PDM) is an emerging research area that aims to integrate prognostic health information and knowledge about the future operating conditions into the process of selecting subsequent actions for the system. The newly developed PDM algorithms require suitable software and hardware platforms for testing under realistic fault scenarios. The paper describes the development of such a platform, based on the K11 planetary rover prototype. A variety of injectable fault modes are being investigated for electrical, mechanical, and power subsystems of the testbed, along with methods for data collection and processing. In addition to the hardware platform, a software simulator with matching capabilities has been developed. The simulator allows for prototyping and initial validation of the algorithms prior to their deployment on the K11. The simulator is also available to the PDM algorithms to assist with the reasoning process. A reference set of diagnostic, prognostic, and decision making algorithms is also described, followed by an overview of the current test scenarios and the results of their execution on the simulator.

Diagnosis and prognosis are necessary tasks for system reconfiguration and fault-adaptive control in complex systems. Diagnosis consists of detec- tion, isolation and identification of faults, while prognosis consists of prediction of the remain- ing useful life of systems. This paper presents an integrated model-based distributed diagnosis and prognosis framework, where system decomposi- tion is used to perform the diagnosis and prog- nosis tasks in a distributed way. We show how different submodels can be automatically con- structed to solve the local diagnosis and prog- nosis problems. We illustrate our approach us- ing a simulated four-wheeled rover for different fault scenarios. Our experiments show that our approach correctly performs fault diagnosis and prognosis in a robust manner.

As software and software intensive systems are becoming increasingly ubiquitous, the impact of failures can be tremendous. In some industries such as aerospace, medical devices, or automotive, such failures can cost lives or endan- ger mission success. Software faults can arise due to the inter- action between the software, the hardware, and the operating environment. Unanticipated environmental changes lead to software anomalies that may have significant impact on the overall success of the mission. Latent coding errors can at any time during system operation trigger faults despite the fact that usually a significant effort has been expended in verification and validation (V&V) of the software system. Nevertheless, it is becoming increasingly more apparent that pre-deployment V&V is not enough to guarantee that a com- plex software system meets all safety, security, and reliabil- ity requirements. Software Health Management (SWHM) is a new field that is concerned with the development of tools and technologies to enable automated detection, diagnosis, prediction, and mitigation of adverse events due to software anomalies, while the system is in operation. The prognos- tic capability of the SWHM to detect and diagnose failures before they happen will yield safer and more dependable systems for the future. This paper addresses the motivation, needs, and requirements of software health management as a new discipline and motivates the need for SWHM in safety critical applications.

PROBABILITY CALIBRATION BY THE MINIMUM AND MAXIMUM PROBABILITY SCORES IN ONE-CLASS BAYES LEARNING FOR ANOMALY DETECTION GUICHONG LI, NATHALIE JAPKOWICZ, IAN HOFFMAN, R. KURT UNGAR ABSTRACT. One-class Bayes learning such as one-class Naïve Bayes and one-class Bayesian Network employs Bayes learning to build a classifier on the positive class only for discriminating the positive class and the negative class. It has been applied to anomaly detection for identifying abnormal behaviors that deviate from normal behaviors. Because one-class Bayes classifiers can produce probability score, which can be used for defining anomaly score for anomaly detection, they are preferable in many practical applications as compared with other one-class learning techniques. However, previously proposed one-class Bayes classifiers might suffer from poor probability estimation when the negative training examples are unavailable. In this paper, we propose a new method to improve the probability estimation. The improved one-class Bayes classifiers can exhibits high performance as compared with previously proposed one-class Bayes classifiers according to our empirical results.