Matthew J. Daigle
I am a research scientist working in the areas of artificial intelligence, machine learning,
and data science. While I have worked in diverse application domains and topics such as electric automobiles and aircraft, robotics, cryogenic propellant systems, simulation, and airspace safety, my main research emphasis is in systems health management, that is, I develop algorithms through which machines can self-diagnose their condition and predict their future failures.
Currently, I am a Principal Data Scientist at NIO USA, Inc.. Prior to joining NIO, I was a Research Computer Scientist and Lead of the Prognostics & Diagnostics Group at NASA Ames Research Center. Prior to that, I was an Associate
Scientist with the University of California, Santa Cruz at NASA Ames.
I received the B.S. degree in Computer Science and Computer and Systems Engineering from Rensselaer Polytechnic Institute, Troy, NY, in 2004, and the M.S. and Ph.D. degrees in Computer Science from Vanderbilt University, Nashville, TN, in 2006 and 2008, respectively. From September 2004 to May 2008, I was a Graduate Research Assistant with the Institute for Software Integrated Systems and Department of Electrical Engineering and Computer Science, Vanderbilt University, Nashville, TN.
Fun Fact: I have an Erdös number of 6, and a Bacon number of 4, resulting in an Erdös-Bacon number of 10.
Email: me at matthewjdaigle.com
Abstract: Prognostics is a systems engineering discipline focused on predicting end-of-life of components and systems. As a relatively new and emerging technology, there are few fielded implementations of prognostics, due in part to practitioners perceiving a large hurdle in developing the models, algorithms, architecture, and integration pieces. Similarly, no open software frameworks for applying prognostics currently exist. This paper introduces the Generic Software Architecture for Prognostics (GSAP), an open-source, cross-platform, object-oriented software framework and support library for creating prognostics applications. GSAP was designed to make prognostics more accessible and enable faster adoption and implementation by industry, by reducing the effort and investment required to develop, test, and deploy prognostics. This paper describes the requirements, design, and testing of GSAP. Additionally, a detailed case study involving battery prognostics demonstrates its use.
- G. Gorospe,
and E. Ng,
“GPU Accelerated Prognostics,” Annual Conference of the Prognostics and Health Management Society 2016, St. Petersburg, FL, October 2017. [show abstract]
Abstract: Prognostic methods enable operators and maintainers to predict future performance for critical systems. However, these methods can be computationally expensive and should be performed each time new information about the system becomes available. In light of these computational requirements, we have investigated the application of graphics processing units (GPUs) as a computational platform for general and real-time prognostics. Recent advances in GPU technology have reduced cost and increased the computational capability of these highly parallel processing units, making them more attractive for the deployment of prognostic software. We present a survey of model-based prognostic algorithms with considerations for leveraging the parallel architecture of the GPU and a case study of GPU-accelerated battery prognostics with computational performance results.
- M. Daigle,
and C. Kulkarni,
“Real-Time Prediction of Safety Margins in the National Airspace,” AIAA Aviation Conference, Denver, CO, June 2017. [show abstract]
Abstract: Underlying all operations in the National Airspace System (NAS) is the concept of safety. Safety, as defined by acceptable levels of risk, is to be maintained at all times. The real-time safety monitoring (RTSM) framework is under development to provide an automated system to quantify safety in the NAS, estimate the current level of safety, and predict the future evolution of safety and the occurrence of events that pose an increased risk to flights so that these occurrences can be managed strategically rather than mitigated reactively. This paper presents the mathematical framework, the models, and the monitoring and prediction algorithms used to achieve this. RTSM computes safety as expressed through a set of safety margins based on user-defined safety metrics, thresolds, and events. Sources of uncertainty are modeled and propagated through the predictions in order to compute the probabilistic evolution of safety and the probability of events that introduce increased risk to operations. A prototype implementation is discussed and results demonstrating feasibility are presented. The results highlight the kinds of predictions that can be computed and the fidelity that is currently achieved.
- L. Spirkovska,
and K. Goebel,
“Real Time Safety Monitoring: Concept for Supporting Safe Flight Operations,” AIAA Aviation Conference, Denver, CO, June 2017. [show abstract]
Abstract: Processes, procedures, regulations, and technologies are continuously evolving to maintain or improve the safety of the National Airspace System (NAS). In this paper, we describe a Real Time Safety Monitoring (RTSM) system that benefits from these efforts to define a set of safety metrics that are automatically monitored in real-time. In addition to providing information about current potentially adverse conditions to a variety of users, from those who need a broad overview of a day's flight operations to those who need to decide on a control tactic to employ in the next five minutes, the RTSM system predicts conditions within a specified prediction horizon. Its intelligent interface alerts the user, presenting the information as appropriate considering the current context and circumstances. We illustrate the system concept with five conceptual use cases, describing which safety metrics may be of the most interest to five user groups and suggesting a multi-modal display format. We posit that having access to information about adverse conditions in time to make efficient preemptive decisions without sacrificing safety will improve the already high level of safety and aid in the expansion planned for the NAS under the Next Generation Air Transportation System (NextGen).
Abstract: Prognostics is the science of making predictions of engineering systems. It is part of a suite of techniques that determine whether a system is behaving within nominal operational bounds and – if it does not – that determine what is wrong and how long it will take until the system no longer fulfills certain functional requirements. This book presents the latest developments and research findings on the topic of prognostics by the Prognostics Center of Excellence at NASA Ames Research Center. The book is intended to provide a practitioner with an understanding of the foundational concepts as well as practical tools to perform prognostics and health management on different types of engineering systems and in particular to predict remaining useful life.
Abstract: This paper presents a computational methodology for uncertainty quantification in predicting the trajectory of a generic, realistic aircraft based on information regarding flight plan, aircraft information, wind and weather information, etc. Predicting the trajectory of aircraft is important from the point of view of analyzing and predicting the safety of the overall airspace, and making risk-informed decisions regarding the operations of the airspace. The proposed methodology is based on using first-principles for analyzing the motion of the aircraft and estimating its future trajectory. Since the core of this problem lies in predicting the future behavior of a generic aircraft, it is essential to understand that it is almost impossible to precisely predict the future trajectory with certainty. Hence, an intuitive approach is to analyze the various sources of uncertainty that affect the aircraft prediction and quantify their combined effect on the whole trajectory. Further, this paper develops a global sensitivity analysis-based methodology to quantify the relative contributions of the various sources of uncertainty to the uncertainty of the overall trajectory. The proposed methodology is illustrated using a numerical example consisting of an aircraft that takes off from the San Francisco International Airport.
Abstract: The operations of a planetary rover depend critically upon the amount of power that can be delivered by its batteries. In order to plan the future operation, it is important to make reliable predictions regarding the end-of-discharge time, which can be used to estimate the remaining driving time and distance. These quantities are stochastic in nature, not only because there are several sources of uncertainty that affect the rover's operation, but also since the future operating conditions cannot be known precisely. This paper presents a computational methodology to predict these stochastic quantities, based on a model of the rover and its batteries. We utilize a model-based prognostics framework that characterizes and incorporates the various sources of uncertainty into these predictions, thereby assisting operational decision-making. We consider two different types of driving scenarios, and develop methods for each to characterize the associated uncertainty. Monte Carlo sampling and the inverse first-order reliability method are used to compute the stochastic predictions of end-of-discharge time, remaining driving time, and remaining driving distance.
Abstract: With increasing complexity of engineering systems, fault diagnostics plays a significant role in ensuring that they operate safely. Such systems most often exhibit mixed discrete and continuous, i.e., hybrid, behavior, and may encounter both parametric faults (unexpected changes in system parameters) as well as discrete faults (unexpected changes in component modes). Diagnosis becomes computationally very complex due to the large number of possible system modes, and possible mode changes that occur near the point of fault occurrence. This paper presents a qualitative fault isolation framework for integrated diagnosis of both parametric and discrete faults in hybrid systems, based on structural model decomposition. Fault isolation is performed by analyzing the qualitative information of the residual deviations, and considering observation delay. The great advantage of structural model decomposition for this problem is that it essentially defines several smaller independent diagnosis problems that become more efficient to solve, and makes the overall diagnosis problem more scalable. To demonstrate and test the validity of our approach, we use a hydraulic multi-tank system as the case study in simulation. Results illustrate that the approach is both efficient and scalable.
Abstract: In the National Airspace System (NAS), safety is assured through a set of rules, regulations, and procedures to respond to unsafe events. However, safety stands to benefit immensely from the introduction of tools and methodologies from Prognostics and Health Management (PHM). PHM will enable the NAS to stochastically predict unsafe states within the NAS, enabling a proactive preventative response strategy, as opposed to a reactive mitigative one. However, current PHM methods do not directly apply to the NAS for several reasons: they typically apply only at the component level, are implemented in a centralized manner, and are focused only on predicting remaining useful life. In this paper, we extend the model-based prognostics approach to PHM in order to provide a framework that can be applied to the NAS. We offer a system-level approach that supports a distributed implementation, and provide algorithms to predict the probability of an unsafe state, either at a specific time or within a time interval, and to predict the time of an unsafe state. Experimental results in simulation demonstrate the new approach.
Abstract: Multiple fault diagnosis is a difficult problem for dynamic systems, and, as a result, most multiple fault diagnosis approaches are restricted to static systems, and most dynamic system diagnosis approaches make the single fault assumption. Within the framework of consistency-based diagnosis, the challenge is to generate conflicts from dynamic signals. For multiple faults, this becomes difficult due to the possibility of fault masking and different relative times of fault occurrence, resulting in many different ways that any given combination of faults can manifest in the observations. In order to address these challenges, we develop a novel multiple fault diagnosis framework for continuous dynamic systems. We construct a qualitative event-based framework, in which discrete qualitative symbols are generated from residual signals. Within this framework, we formulate an online diagnosis approach and establish definitions of multiple fault diagnosability. Residual generators are constructed based on structural model decomposition, which, as we demonstrate, has the effect of reducing the impact of fault masking by decoupling faults from residuals, thus improving diagnosability and fault isolation performance. Through simulation-based multiple fault diagnosis experiments, we demonstrate and validate the concepts developed here, using a multi-tank system as a case study.