Woosun An

Data Visualization, Data Reduction and Classifier Fusion for Intelligent Fault Diagnosis in Gas Turbine Engines

In this paper, we investigate four key issues associated with data-driven approaches for fault classification using the Pratt and Whitney commercial dual-spool turbofan engine data as a test case. The four issues considered here include the following. (1) Can we characterize, a priori, the difficulty of fault classification via self-organizing maps? (2) Do data reduction techniques improve fault classification performance and enable the implementation of data-driven classification techniques in memory-constrained digital electronic control units (DECUs)? (3) When does adaptive boosting, an incremental fusion method that successively combines moderately inaccurate classifiers into accurate ones, help improve classification performance? (4) How to synthesize classifier fusion architectures to improve the overall diagnostic accuracy? The classifiers studied in this paper are the support vector machine, probabilistic neural network, k-nearest neighbor, principal component analysis, Gaussian mixture models, and a physics-based single fault isolator. As these algorithms operate on large volumes of data and are generally computationally expensive, we reduce the data set using the multiway partial least squares method. This has the added benefits of improved diagnostic accuracy and smaller memory requirements. The performance of the moderately inaccurate classifiers is improved using adaptive boosting. These results are compared to the results of the classifiers alone, as well as different fusion architectures. We show that fusion reduces the variability in diagnostic accuracy, and is most useful when combining moderately inaccurate classifiers.

Quantifying the Impact of Information and Organizational Structures via Distributed Auction Algorithm: Point-to-Point Communication Structure

This paper presents how information and organizational structures with point-to-point communication structure impact team coordination in a distributed task-asset allocation problem. A key distinguishing characteristic of this problem is that each decision maker knows only a part of the weight matrix and/or controls a subset of the assets. Here, we extend the distributed algorithm developed for blackboard communication structure in another part of this work to the point-to-point communication structure. Our results indicate that edge organizations with horizontal and vertical information structures exhibit shorter delays than those with block diagonal and checkerboard information structures. We also showed how our findings can be applied effectively to mission planning of the Navys maritime operations center.

Data Visualization, Data Reduction and Classifier Fusion for Intelligent Fault Detection and Diagnosis in Gas Turbine Engines

In this paper, we investigate four key issues associated with data-driven approaches for fault classification using the Pratt and Whitney commercial dual-spool turbofan engine data as a test case. The four issues considered here include: (1) Can we characterize, a priori, the difficulty of fault classification via self-organizing maps? (2) Do data reduction techniques improve fault classification performance and enable the implementation of data-driven classification techniques in memory-constrained digital electronic control units (DECUs)?, (3) When does adaptive boosting, an incremental fusion method that successively combines moderately inaccurate classifiers into accurate ones, help improve classification performance?, and (4) How to synthesize classifier fusion architectures to improve the overall diagnostic accuracy? The classifiers studied in this paper are the support vector machine (SVM), probabilistic neural network (PNN), k-nearest neighbor (KNN), principal component analysis (PCA), Gaussian mixture models (GMM), and a physics-based single fault isolator (SFI). As these algorithms operate on large volumes of data and are generally computationally expensive, we reduce the dataset using the multi-way partial least squares (MPLS) method. This has the added benefits of improved diagnostic accuracy and smaller memory requirements. The performance of the moderately inaccurate classifiers is improved using adaptive boosting (AdaBoost). These results are compared to the results of the classifiers alone, as well as different fusion architectures. We show that fusion reduces the variability in diagnostic accuracy, and is most useful when combining moderately inaccurate classifiers.

Hidden Markov Model and Auction-Based Formulations of Sensor Coordination Mechanisms in Dynamic Task Environments

In this paper, multistage auction-based intelligence, surveillance, and reconnaissance (ISR) sensor coordination mechanisms are investigated in the context of dynamic and uncertain mission environments such as those faced by expeditionary strike groups. Each attribute of the mission task is modeled using a hidden Markov model (HMM) with controllable emission matrices, corresponding to each ISR asset package (subset of sensors). For each HMM-asset package pair, we evaluate a matrix of information gains (uncertainty reduction measures). The elements of this matrix depend on the asset coordination structure and the concomitant delays accrued. We consider three coordination structures (distributed ISR coordination, ISR officer serving as a coordinator, and ISR officer serving as a commander) here. We evaluate these structures on a hypothetical mission scenario that requires the monitoring of ISR activities in multiple geographic regions. The three structures are evaluated by comparing the task state estimation error cost, as well as travel, waiting, and assignment delays. The results of the analysis were used as a guide in the design of a mission scenario and asset composition for a team-in-the-loop experimentation. Our solution has the potential to be a mixed initiative decision support tool to an ISR coordinator/commander, where the human provides possible ISR asset package-task pairings and the tool evaluates the efficacy of the assignment in terms of task accuracy and delays. We also apply our approach to a hypothetical disaster management scenario involving chemical contamination and discuss the computational complexity of our approach.

Quantifying the impact of information and communication structures via distributed auction algorithm

Task-asset assignment is a fundamental problem paradigm in a wide variety of applications. A typical problem scenario involves a single decision maker (DM) who has complete knowledge of the weight (or reward/benefit/accuracy) matrix and who can control any of the assets to execute the tasks. Motivate by planning problems arising in distributed organizations, this paper introduces a novel variation of the assignment problem, wherein there are multiple DMs and each DM know only a part of the weight matrix and/or controls a subset of the assets. We extend the auction algorithm to such realistic settings with various partial information structures and communication structures. We show that by communicating the bid, the best and the second best profits among DMs and with a coordinator, the DMs can reconstruct the centralized assignment solution. The auction setup provides a nice analytical framework for formalizing how team members build internal models of other DMs and achieve team cohesiveness over time.

Decision support software for Anti-Submarine warfare mission planning within a dynamic environmental context

Anti-Submarine Warfare (ASW) involves effective allocation and path planning of ASW platforms to search for, detect, classify, track and prosecute hostile submarines within an evolving environment. As the environmental context evolves rapidly, continuously collected Meteorological and Oceanographic (METOC) data is used for assessing the impact of the current and forecasted environment on individual sensors and weapon platforms, as well as on tactics in the form of performance surfaces, which is presented to the commanders in making go/no-go decisions. However, due to the overwhelming amount of METOC information, it is very challenging for the commanders to interpret and analyze the data for generating plans or evaluating courses of action in a timely manner. In this paper, motivated by the need to assist ASW commanders in making proactive decisions in an evolving environmental context, we present a decision support tool for modeling and incorporating the appropriate METOC information from multiple sources and further utilizing it to determine the search regions and optimal trajectories to search for and track the enemy submarines in a timely manner.

Decision support information integration platform for context-driven interdiction operations in counter-smuggling missions

Context-driven decision making is at the top of the Navys agenda of important concepts to be embedded in future proactive decision support systems for command decision making. There are manifold challenges associated with relaying contextual data in a timely manner to the decision maker. In the counter-smuggling domain, for example, although high value information is accessible, it is dispersed across databases and the decision making team. There are numerous research challenges in integrating this information in an efficient manner to effectively present viable courses of action to a decision making team. In this paper, we propose a decision support tool for counter-smuggling missions modeled as a stochastic control problem of dynamically managing assets to maximize the probability of detecting and interdicting maritime illicit trafficking operations. We additionally propose a method to explain the algorithm behavior to the human decision maker and provide them with interactive controls to develop “what-if” solutions or to constrain solutions to a desired path.

Context-Aware Decision Support for Anti-Submarine Warfare Mission Planning Within a Dynamic Environment

Anti-submarine warfare (ASW) missions are the linchpin of maritime operations involving effective allocation and path planning of scarce assets to search for, detect, classify, track, and prosecute hostile submarines within a dynamic and uncertain mission environment. Motivated by the need to assist ASW commanders to make better decisions within an evolving mission context, we investigate a moving target search problem with multiple searchers and develop a context-driven decision support tool for the ASW mission planning problem. Given the spatial probability distribution of a target submarine, sensor detection probability surfaces from meteorological and oceanographic products, and the risk to the fleet as a function of distance of the target from the fleet, we model and formulate the ASW asset allocation and search path planning problem using a hidden Markov modeling framework. We propose a two phase approach to solve this NP-hard problem. In phase I, we partition the geographic area, satisfying contiguity constraints, into search regions using an evolutionary algorithm (EA) coupled with a Voronoi tessellation approach, and allocate the assets to partitioned search areas using the auction algorithm. In phase II, we construct a dynamic search plan for each asset over the search interval using EA. We evaluate our approach via a hypothetical ASW scenario to monitor an enemy submarine in a geographic region via multiple assets. We compare our results to various search path planning strategies that, using the context-driven decision support tool developed here, revise the search regions at periodic intervals given a fixed total search time.

Rollout strategy for Hidden Markov Model (HMM)-based dynamic sensor scheduling

In this paper, a hidden Markov model (HMM)-based dynamic sensor scheduling problem is formulated, and solved using rollout concepts to overcome the computational intractability of the dynamic programming (DP) recursion. The problem considered here involves dynamically sequencing a set of sensors to minimize the sum of sensor cost and the HMM state estimation error cost. The surveillance task is modeled as a single HMM with multiple emission matrices corresponding to each of the sensors. The rollout information gain (RIG) algorithm proposed herein employs the information gain (IG) heuristic as the base algorithm. The RIG algorithm is illustrated on an intelligence, surveillance, and reconnaissance (ISR) scenario of a village for the presence of weapons and terrorists/refugees. Extension of the RIG strategy to monitor multiple HMMs involves combining the information gain heuristic with the auction algorithm that computes the K -best assignments at each decision epoch of rollout.