Pritthi Chattopadhyay

Context-aware Dynamic Data-driven Pattern Classification ∗

This work aims to mathematically formalize the notion of context, with the purpose of allowing contextual decision-making in order to improve performance in dynamic data driven classification systems. We present definitions for both intrinsic context, i.e. factors which directly affect sensor measurements for a given event, as well as extrinsic context, i.e. factors which do not affect the sensor measurements directly, but do affect the interpretation of collected data. Supervised and unsupervised modeling techniques to derive context and context labels from sensor data are formulated. Here, supervised modeling incorporates the a priori known factors affecting the sensing modalities, while unsupervised modeling autonomously discovers the structure of those factors in sensor data. Context-aware event classification algorithms are developed by adapting the classification boundaries, dependent on the current operational context. Improvements in context-aware classification have been quantified and validated in an unattended sensor-fence application for US Border Monitoring. Field data, collected with seismic sensors on different ground types, are analyzed in order to classify two types of walking across the border, namely, normal and stealthy. The classification is shown to be strongly dependent on the context (specifically, soil type: gravel or moist soil).

Path planning in GPS-denied environments via collective intelligence of distributed sensor networks

ABSTRACT This paper proposes a framework for reactive goal-directed navigation without global positioning facilities in unknown dynamic environments. A mobile sensor network is used for localising regions of interest for path planning of an autonomous mobile robot. The underlying theory is an extension of a generalised gossip algorithm that has been recently developed in a language-measure-theoretic setting. The algorithm has been used to propagate local decisions of target detection over a mobile sensor network and thus, it generates a belief map for the detected target over the network. In this setting, an autonomous mobile robot may communicate only with a few mobile sensing nodes in its own neighbourhood and localise itself relative to the communicating nodes with bounded uncertainties. The robot makes use of the knowledge based on the belief of the mobile sensors to generate a sequence of way-points, leading to a possible goal. The estimated way-points are used by a sampling-based motion planning algorithm to generate feasible trajectories for the robot. The proposed concept has been validated by numerical simulation on a mobile sensor network test-bed and a Dubin’s car-like robot.

Path planning in GPS-denied environments: A collective intelligence approach

This paper proposes a framework for reactive goal-directed navigation without global positioning facilities in unknown environments. A mobile sensor network is used for localization of regions of interest for path planning of an autonomous mobile robot in the absence of global positioning facilities. The underlying theory is an extension of a generalized gossip algorithm that has been recently developed in a language-measure-theoretic setting. The gossip algorithm has been used to propagate local decisions of target detection over a mobile sensor network and thus, it generates a belief for the target detected over the network. The proposed concept has been validated through numerical experiments with a mobile sensor network and a point mass robot.

Performance robustness of feature extraction for target detection & classification

Performance robustness of feature extraction with respect to environmental uncertainties is often critical for automated target detection & classification. This paper focuses on performance robustness in the sense that the extracted features are desired to be largely insensitive to environmental uncertainties, while they should be capable of recognizing the effects of small perturbations in the underlying system dynamics for detection & classification. From this perspective, performance robustness of three feature extraction algorithms, namely, principal component analysis, cepstrum, and symbolic dynamic filtering, is evaluated for target classification by making use of the respective field data collected from different sites. These algorithms have been evaluated for robust classification of two different types of mortar launchers with acoustic sensing systems, based on the training and testing data sets from the same and different field sites. The results, generated with training and testing data from different field sites, characterize performance robustness of the respective feature extraction algorithms, when compared with those generated with the corresponding data sets from the same field site.

Dynamic Data-Driven Design of Lean Premixed Combustors for Thermoacoustically Stable Operations

Prediction of thermoacoustic instabilities is a critical issue for both design and operation of combustion systems. Sustained high-amplitude pressure and temperature oscillations may cause stresses in structural components of the combustor, leading to thermomechanical damage. Therefore, the design of combustion systems must take into account the dynamic characteristics of thermoacoustic instabilities in the combustor. From this perspective, there needs to be a procedure, in the design process, to recognize the operating conditions (or parameters) that could lead to such thermoacoustic instabilities. However, often the available experimental data are limited and may not provide a complete map of the stability region(s) over the entire range of operations. To address this issue, a Bayesian nonparametric method has been adopted in this paper. By making use of limited experimental data, the proposed design method determines a mapping from a set of operating conditions to that of stability regions in the combustion system. This map is designed to be capable of (i) predicting the system response of the combustor at operating conditions at which experimental data are unavailable and (ii) statistically quantifying the uncertainties in the estimated parameters. With the ensemble of information thus gained about the system response at different operating points, the key design parameters of the combustor system can be identified; such a design would be statistically significant for satisfying the system specifications. The proposed method has been validated with experimental data of pressure time-series from a laboratory-scale lean-premixed swirlstabilized combustor apparatus. [DOI: 10.1115/1.4037307]

Real-time identification of state-of-charge in battery systems: Dynamic data-driven estimation with limited window length

This paper presents a symbolic dynamic method for real-time estimation of battery state-of-charge (SOC). In the proposed method, symbol strings are generated by partitioning (finite-length) time windows of synchronized input-output (e.g., current-voltage) pairs in the respective two-dimensional space. Then, a special class of probabilistic finite state automata (PFSA), called D-Markov machine, is constructed from the symbol strings to extract pertinent features. The SOC estimation is formulated as a sequential estimation scheme with adaptive acceptance of new features to circumvent the problem of having potential outliers. A major challenge is that SOC value is continuously varying during the operation. While modeling and analysis of such time-varying problems is computationally intensive, the data-driven approach requires adequate length of time series data for statistically significant analysis. From these perspectives, a critical aspect is to determine an optimal (or suboptimal) length of the analysis window to make a tradeoff between estimation accuracy and dynamic sensitivity. The proposed method has been validated on experimental data of a commercial-scale lead-acid battery.

Simultaneous tracking and counting of targets in a sensor network

Unattended ground sensors (UGS) are widely used to monitor human activities, such as pedestrian motion and detection of intruders in a secure region. This paper presents an algorithm for counting and tracking humans moving through a UGS network. Each node of this sensor network is equipped with a geophone (i.e., seismic sensor) and a microphone (i.e., acoustic sensor). The proposed method analyzes the relational dependence among the responses of sensors at various nodes as the targets walk through the network. The energy distribution across the network for different number of targets walking at different distances from the nodes has been analyzed to predict the number and location of targets in the sensor network field. The proposed concept has the advantages of having fast execution time and low memory requirements and is potentially well-suited for real-time implementation on in-situ computational platforms. Keywords—Personnel detection, seismic sensing, acoustic sensing, sensor-network-based fusion.

Identification of battery parameters via symbolic input-output analysis: A dynamic data-driven approach

This paper presents real-time parameter identification in battery systems as a paradigm of dynamic data-driven application systems (DDDAS). In the proposed method, symbol sequences are generated by partitioning (finite-length) time series data of synchronized input-output (i.e., current-voltage) pairs in the respective two-dimensional space. Then, a special class of probabilistic finite state automata (PFSA), called D-Markov machine, is constructed to extract pertinent features from the statistics of time series as state probability vectors. The proposed method has been validated on (approximately periodic) experimental data of a lead-acid battery for real-time identification of its pertinent parameters: State-of-Charge (SOC) and State-of-Health (SOH). The results of experimentation show that the analysis of input-output-pair data exceeds the performance of output-only data analysis.

Dynamic Data-Driven Combustor Design for Mitigation of Thermoacoustic Instabilities

A critical issue in design and operation of combustors in gas turbine engines is mitigation of thermoacoustic instabilities, because such instabilities may cause severe damage to the mechanical structure of the combustor. Hence, it is important to quantitatively assimilate the knowledge of the system conditions that would potentially lead to these instabilities. This technical brief proposes a dynamic data-driven technique for design of combustion systems by taking stability of pressure oscillations into consideration. Given appropriate experimental data at selected operating conditions, the proposed design methodology determines a mapping from a set of operating conditions to a set of quantified stability conditions for pressure oscillations. This mapping is then used as an extrapolation tool for predicting the system stability for other conditions for which experiments have not been conducted. Salient properties of the proposed design methodology are: (1) It is dynamic in the sense that no fixed model structure needs to be assumed, and a suboptimal model (under specified user-selected constraints) is identified for each operating condition. An information-theoretic measure is then used for performance comparison among different models of varying structures and/or parameters and (2) It quantifies a (statistical) confidence level in the estimate of system stability for an unobserved operating condition by using a Bayesian nonparametric technique. The proposed design methodology has been validated with experimental data of pressure time-series, acquired from a laboratory-scale leanpremixed swirl-stabilized combustor. [DOI: 10.1115/1.4040210]