Search NASA⌕ Search

Engineering topics

Hannah S Walsh

Publications and source records attributed to Hannah S Walsh.

12 records
All sourcesNASA NTRSDOE OSTI
NASA NTRS2023

BERT-Based Topic Modeling and Information Retrieval to Support Fishbone Diagramming for Safe Integration of Unmanned Aircraft Systems in Wildfire Response

Recent concepts for emerging wildfire response operations have included unmanned aircraft systems (UAS) due to their increasing accessibility and capabilities. To integrate UAS into wildfire response safely, researchers have studied the use of large repositories of historic incident reports to improve the scope of root cause analysis. Recent work has emphasized applying state-of-the-art natural language processing techniques to extract useful information from these repositories. However, it has not yet been studied how these results can be interpreted and integrated into the systems engineering process. In this work, we propose a process in which Bidirectional Encoder Representations from Transformers (BERT)-based topic modeling and information retrieval are applied to a relevant set of documents in order to support the development of a fishbone diagram in a semiautomated process. High-level themes in the document set are identified using topic modeling, which are then refined and interpreted by a human analyst. Then, the themes are used to guide a finer search using information retrieval, which returns specific incident reports of relevance. This provides traceability to specific incidents as well as broader categorizations that comprise the fishbone branches. We apply the proposed process to relevant documents from NASA’s Aviation Safety Reporting System (ASRS). The proposed process is widely applicable when relevant documents are available, and the results from this study will be useful to identifying potential causes of wildfire response UAS incidents.

Hazard analysis↗
NASA NTRS2023

SafeAeroBERT: Towards a Safety-Informed Aerospace-Specific Language Model

As aviation systems continue to operate with high traffic, large amounts of documents containing safety-relevant data continue to be generated via reporting systems such as the ASRS. Advanced natural language processing techniques, specifically pre-trained language models, have shown great success in domain-specific applications; however, the text in aviation safety reports is inundated with jargon and thus not fully utilized by general pre-trained models. In this research, we work towards developing a safety-informed aerospace-specific language model by pre-training a Bidirectional Encoder Representations from Transformer (BERT) model on reports from the Aviation Safety Reporting System and the National Transportation Safety Board. The resulting model, called SafeAeroBERT, is fine-tuned for the specific task of document classification, and can be further tuned for named-entity recognition, relation detection, information retrieval, and summarization. Results from the classification task are compared between SafeAeroBERT, the base BERT, and SciBERT models and show SafeAeroBERT outperforms the general BERT and SciBERT on classifying reports about human factors, aircraft, and procedure. SafeAeroBERT can be used on custom tasks, not limited to document classification, and is intended to aid an intelligent knowledge manager for safety report repositories.

Aviation↗
NASA NTRS2022

Supporting Hazard Analysis for Wildfire Response Using fmdtools and MIKA

The System Wide Safety (SWS) Safety Demonstrator (SD) Series drives development of an increasingly capable In-Time Aviation Safety Management System (IASMS) focusing on humanitarian applications, starting with wildfire response (SD-1). The goals of this report are to (1) provide an early hazard analysis and mitigation evaluation of wildfire response to support these efforts and (2) provide a demonstration of capabilities of the Fault Model Design Tools (fmdtools) and Manager for Intelligent Knowledge Access (MIKA) tools. fmdtools provides a modeling, simulation, and resiliency analysis framework in which a wildfire response model, the System Modeling and Analysis of Resiliency in Scalable Traffic Management for Emergency Response Operations (SMARt-STEReO), is built. MIKA is an intelligent knowledge manager with several capabilities, including assisting in hazard analysis by extracting and analyzing hazards from historical incident reports. The following topics are covered in the report: Understanding Wildfire Hazard Dynamics. We provide a description and simulated examples of how hazards occur in the SMARt-STEReO model of wildfire response and their effect on its outcome. This provides a common mental model and focuses the analysis presented in the remainder of the report. Wildfire Hazard Identification. MIKA identifies wildfire hazards from three relevant datasets: the ICS-209-PLUS, SAFECOM, and SAFENET. Hazards are manually organized into a taxonomy and MIKA analyzes each hazard’s effects, likelihood, severity, and risk. Evaluating Mitigation Strategies. The SMARt-STEReO wildfire response model built in fmdtools evaluates a subset of identified hazards. Specifically, we simulate the effect of communications faults and equipment faults on operator safety, the effect of changing winds and flammability, and a scenario with multiple ignition points and heavy smoke. Tool Limitations and Usage Considerations. We provide a discussion of appropriate tool use cases as well as limitations and considerations for usage. The tool findings are used to synthesize recommendations for wildfire response operations, which can be captured as part of an IASMS. Key recommendations are as follows: Hazards are identified from a broad spectrum of sources including aircraft subsystems, operational sources, and ground crew operations. Highest risk operational environment hazards identified are Evacuations. The highest risk manned aerial operations hazard categorized is Jumper Operations Mishap. Ground crew hazards that are highest risk are Burns, Cargo Operations Overhead, Dehydration, Entrapment, Falling Objects, Heart Attacks, Heat Exhaustion, Inadequate Training or Certification, Vehicle Breakdown, and Vehicle Collision. Modelled containment failures arise from a mismatch between the difficulty of the firefighting scenario and the capacity (e.g., speed, effectiveness, awareness) of the response. In firefighting scenarios where containment is possible (e.g., because the fire does not spread too quickly), these mismatches can occur because of a change in environmental conditions (e.g., wind, flammability, etc) or because of planning, equipment, or communications faults. Improvements to communications increase the capacity of the firefighting response by reducing the time needed to respond to the fire. While surveillance does not increase this capacity by itself, it increases operator safety by increasing state awareness, enabling firefighters to evade approaching fires. Increasing both has a synergistic effect. In general, these performance and resilience increases generalize over fault scenarios as well as unforeseen changes to circumstances (i.e., wind, aridity, etc.). However, these improvements need to be designed so as not to make the system prone to persistent large-scale communications outages, which can reduce performance.

Hazard analysis↗
NASA NTRS2020

SMARt-STEReO: Preliminary Model Description

Wildfires have increasingly become major threat to US towns and cities, with California alone having its most destructive wildfire seasons to date in 2017 and 2018 and 2020 already becoming one of the worst fire years on record, causing the large-scale evacuations, the destruction of towns and property, and a significant release of hazardous smoke over the entire west coast. Thus, the U.S. Forest Service recently spent over 50% of its budget on wildfire management – funds which could otherwise be directed to science that could support fire prevention. Aerial support plays a major role in fighting wildfires. Airtankers, helicopters, and other aerial assets support the construction of fire-lines, gather data, and transport crews and equipment. However, presently, aerial firefighting operations rely on relatively unsophisticated technologies for communications, such as over-the-air radios, which limit the ability of pilots and management to relay fire data and coordinate operations. Aerial operations in and around fires are additionally high-risk activities due to the variable, dangerous conditions and complex, technically difficult maneuvers which must be performed to, for example, conduct a retardant drop. Thus, between 2000 and 2013, of the 298 wildland firefighter fatalities, 26% were related to aerial operations. As a result, there is significant potential to both improve the performance and resiliency of wildfire response while reducing risk to human operators.

Daniel E Hulse↗
NASA NTRSDate not supplied

Detecting And Characterizing Archetypes of Unintended Consequences in Engineered Systems

When designing engineered systems, the potential for unintended consequences of design policies or design decisions exists despite best intentions. Conditions that might cause the formation of unintended consequences are often known only in hindsight. However, since these conditions are associated with a single event, it is difficult to uncover the general patterns of conditions leading to unintended consequences. In this research, patterns of conditions associated with unintended consequences are learned from historical data and represented in the form of archetypes. While previous work using systems theoretic modeling has identified high-level archetypes, this work leverages a self-organizing map to learn archetypes of unintended consequences from human-tagged risk factors in a large data set of lessons learned from adverse events at NASA. The sixty-six identified archetypes contain patterns of conditions such as complexity and human-machine interaction associated with the formation of unintended consequences. To validate the archetypes, a sample of the archetypes is represented using system dynamics in order to illustrate that the identified archetypes are specialized versions of known high-level archetypes of unintended consequences. While the research is based upon a specific dataset, the archetypes apply to any engineered system and the pattern of leading indicators open a new path to manage unintended consequences and mitigate the magnitude of potentially adverse outcomes.

Hannah S Walsh↗
NASA NTRSDate not supplied

Knowledge Discovery for Early Failure Assessment of Complex Engineered Systems Using Natural Language Processing

Emerging complex engineered systems may have unexpected safety issues due to novel operational environments, increasing autonomy, human-machine interaction, and other factors. To prevent failures in operation or testing that necessitate costly redesign, it is desirable to predict likely failure modes early in the design process. Information about past engineering failures in natural language format presents one possible solution by enabling the retrieval of information that can inform new designs. However, identifying documents containing usable information and extracting the required information can be prohibitively time-consuming when implemented at scale. In this research, an automated natural language processing (NLP) framework is proposed to discover relevant knowledge from documents containing failure-related design information. The framework is applied to NASA’s Lessons Learned Information System (LLIS),which is publicly available. Documents containing usable information are filtered using two different NLP-based models. Next, from the identified usable documents, a failure taxonomy is extracted using a partitioned hierarchical topic modeling approach. Partitions of the document describe different sections of the failure taxonomy – i.e., failure, cause of failure, and recommendations – as indicated by the structure of the original document. The extracted failure taxonomy can be leveraged in early design failure assessment methods. Moreover, the framework can be used to identify documents containing usable failure-related design information from other databases and extract relevant information from these documents.

Documentation and Information Science↗
NASA NTRSDate not supplied

Machine Learning Enabled Quantitative Risk Assessment of Aerial Wildfire Response

Aerial wildfire operations are high risk and account for a large number of firefighter deaths. Increasing intensity of wildfires is driving a surge in aerial operations, while simultaneously there is growing interest in improving system safety and performance. In this work, wildfire aviation mishaps documented using the SAFECOM system are analyzed using a previously developed framework for hazard extraction and analysis of trends (HEAT). Hazards and specific failure modes are extracted from the narrative data in SAFECOM forms using natural language processing techniques. Metrics for each hazard are calculated, including frequency, rate, and severity. We examine whether these metrics change over time, and whether they are related to metadata, such as region and aircraft type. The results of the hazard analysis are presented in a risk matrix, identifying the highest and lowest risk hazards based on rate of occurrence and average severity. Results identify jumper operations hazards as high-risk, in addition to bucket drop failures, cargo let down failures, and severe weather as medium risk.

machine learning↗
NASA NTRSDate not supplied

Open Data Integration (ODIN): A Concurrent, Distributed Message-Based Architecture and Framework for Disaster Response

The Runtime for Airspace Concept Evaluation (RACE) is an open-source software architecture and framework to build configurable, highly concurrent and distributed message-based systems that offer scalable, low-latency performance on commodity hardware. RACE was used in commercial aviation applications to rapidly build systems that span several machines (including synchronized displays), interface existing hardware simulators and other live data feeds, and incorporate sophisticated visualization components such as NASA WorldWind. These RACE applications validated elements of the FAA’s System Wide Information Management (SWIM) Program, handling up to 1000 messages/sec from diverse sources (SFDPS, TFM-DATA, TAIS, ASDE-X, ITWS and local ADS) for 4,500 simultaneous flights tracked in the next-generation air transportation system’s digital backbone. We have since generalized RACE to support Open Data Integration (ODIN) applications outside aviation. Systems built with RACE/ODIN can be deployed in the field, on commodity hardware, and operate with limited or intermittent connectivity to the outside world. Our primary use case is a web-server with local/persistent data storage that runs within and only serves the stakeholder network (e.g. an incident command post). We are tailoring the RACE/ODIN system to support wildland fire management for the upcoming NASA Wildland Fire Safety Demonstration Series. RACE-ODIN is under consideration for application in the Scalable Traffic Management for Emergency Response Operations project, or STEReO, which aims to create a system that can be deployed during emergencies, to coordinate multiple elements of disaster response. Such data sources predominantly come from existing services on the internet (e.g. weather and satellite data, imported from so called "edge servers") but can also include dynamic (real-time) data from computer simulations and within the stakeholder network (such as aircraft and personnel tracking information). We will present the architecture and ODIN system demonstration incorporating local data from instrumented power-line towers, interpolated weather data and geospatial data from space-based platforms.

Joseph C Coughlan↗
NASA NTRSDate not supplied

What Went Wrong: A Survey of Wildfire UAS Mishaps through Named Entity Recognition

Increasingly, unmanned aircraft systems (UAS) are being applied to wildfire incidents for tasks such as mapping, aerial ignition, and delivery. As a result, aviation incident reporting systems for wildfires are beginning to accumulate data related to UAS mishaps in wildfire response. In this research, we apply state-of-the-art natural language processing (NLP) techniques to develop a custom Named Entity Recognition (NER) model which extracts entities relevant to safety analysts. The custom NER model is built by fine-tuning an existing Bidirectional Encoder Representations from Transformers (BERT) model, resulting in a generalizable NER model that can extract engineering relevant entities including failure modes, causes, effects, control processes, and recommendations from failure-relevant text. This model performs passably, with a weighted average f1 score of 0.33 across entity types, indicating more labeled training data is needed. Extracted entities are used to form a Failure Modes and Effects Analysis (FMEA)-style survey of wildfire UAS mishaps reported using the SAFECOM system. Similar mishaps are manually clustered and reported as single rows within an FMEA. Foreach cluster, we compute frequency, severity, and overall riskin accordance with FAA standards. This methodology can beapplied as part of a broader safety management system totrack trends in mishaps (e.g., likelihood, severity) and discoverknowledge (e.g., causes, effects) that can be utilized to improvesafety outcomes and system performance.

Machine Learning↗
NASA NTRSDate not supplied

From BERTopic to SysML: Informing Model-Based Failure Analysis With Natural Language Processing for Complex Aerospace Systems

The development of emerging complex aerospace systems will require new approaches for capturing safety incident scenarios as early as possible in the design phase. However, for novel systems, relevant data available is limited. In this work, we propose a framework informing model-based mission assurance activities with historical incident reports, lessons learned, or other relevant engineering documents using natural language processing. In doing so, we investigate whether there is useful information in data sets that are relevant, if not identical, to the system under design and whether, through rigorous systems engineering practice, this information can be effectively leveraged through model-based failure analysis. In a worked case study, we apply state-of-the-art topic modeling techniques to two data sets, a mission relevant data set and a system relevant data set. The sets of topics are merged and interpreted to form a preliminary list of failure topics that can be used to inform the identification of off-nominal modes in the model-based failure modes and effects analysis development. Once data from the system in operation is available, it can be used to update the topics identified. By extracting information about likely failures from relevant historical data sets and utilizing model-based mission assurance to ensure relevance and rigor, unanticipated failures can be reduced, and projects can more effectively learn from past missions.

Failure Analysis↗
NASA NTRSDate not supplied

Towards Computational Functional Hazard Assessment (CFHA): A Gap Analysis and Concept for Emerging Aviation Systems

Given the current evolution of the National Airspace and future trajectory towards novel and evolving operations with varying levels of autonomy, complexity, and acceptable risk, there is an opportunity to support safety assurance by extending existing methodologies, such as Functional Hazard Assessment (FHA). In response to challenges in performing FHA for novel aviation concepts, we propose a concept for Computational Functional Hazard Assessment (CFHA), which provides processes, methods, and tools for incorporating external data to facilitate further exploration of the hazard space iterativelty throughout the design process. The core components of CFHA involve knowledge capture from historical and operational data, functional architecture specification via a formal modeling language, and simulation for hazardous scenario analysis. Through this concept, we aim to adapt conventional safety assessment to address the increasingly complex hazard space generated from emerging operations.

Seydou Mbaye↗