Technical Session Abstracts

1A1: A Decision-Based Mobility Model for Semi and Fully Autonomous Vehicles

Contributors: Vijitashwa Pandey, Line Deschenes and Sam Kassoumeh (Oakland University)

The objective of this project is to rigorously quantify the mobility of ground vehicles exhibiting partial to full autonomy, so that acquisition and operational decisions regarding them can be made. In this talk, we will present the approach and the techniques we have developed to accomplish this objective. Multiattribute utility models are used to encode decision maker preferences over mission attributes such as navigation error and travel time, among others. There are many relevant uncertainties that affect these attributes, such as in soil characteristics and the quality of communication links. Frequently these uncertainties can be reduced or eliminated with the collection of data. We investigate how to evaluate the value of collecting such information, since data collection such as through experiments or remote sensing, can be expensive and time consuming. Value of information studies can help decide when information collection is worthwhile. Statistical techniques such as Metropolis-Hastings algorithm can be used to efficiently generate samples from posterior distributions, when new information becomes available. New analytical and simulation based methods to evaluate full-attribute and attribute-wise value of information are also being investigated. A Matlab based software tool is also under development to perform the above analyses.

1A2: Adversarially Robust Coordination for Autonomous Multi-Vehicle Systems

Contributors: Dimitra Panagou (PI), James Usevitch (PhD Candidate), Sahib Dhanjal (MS Student), Aishwarya Unnikrishnan (MS Student) (University of Michigan); Paramsothy Jayakumar, Paul Muench (GVSC); John Sauter (Soartech)

This ARC project aims to increase the existing levels of autonomy of networked multi-vehicle systems in adversarial environments by considering both the safety and the resilience aspects of dynamic multi-agent networks. Safety refers to the generation of guaranteed collision-free trajectories for multiple vehicles (agents) so that they navigate efficiently in cluttered environments while collaborating towards a common task (e.g., formation control). Resilience refers to the guaranteed safe accomplishment of the mission despite the presence of possible adversaries that send malicious data over compromised communication links. In this talk we will present our results towards resilient communication structures and adversarially robust coordination mechanisms that maintain safety of the individual agents, filter out the effects of adversarial agents, and guarantee mission accomplishment.

1A3: AI-Based Attacker-Defender Dynamics of Adaptable Fleets of Autonomous Vehicles

Contributors: Xingyu (Gavin) Li, Panos Papalambros (Co-PI), Bogdan I. Epureanu (PI) (University of Michigan)l Matthew Castanier, Richard Gerth (GVSC)

Fast-paced changes in combat environments and technological developments are increasing ground vehicles. Efforts of the Army Futures Command are devoted to addressing this issue. One possible solution is to enhance the modularity of future combat vehicles. Modular vehicles can be built with interchangeable substantial components, i.e. modules which are easily of assembled and disassembled by relatively simple means. Such modular characteristics allow acquisition cost reductions and also adaptability through real-time fleet reconfiguration to meet dynamic demands. These demands are time-varying and highly stochastic because they are created in part by adversarial actions. To capture these characteristics, we developed a synthetic environment where we create an intelligent agent-based model to imitate the decision making process during fleet operations, which combines real-time optimization with artificial intelligence. Agents are capable to infer future adversarial actions based on historical data and to optimize dispatch and operation decisions accordingly. We simulate an attacker-defender game in this environment between two adversarial and intelligent fleets, one modular and the other conventional. Further, we implement data mining techniques to extract heuristics, and implement algorithms for fleet management by studying the optimized operations strategies from the simulations of attacker-defender games in different mission scenarios. Given the same level of resources and intelligence, we explore the characteristics of the logistic burden and the benefits of the adaptability of the modular fleet while highlighting tactical advantages in terms of win rate, unpredictability and damage suffered.

1A4: Trust-based Control and Scheduling for UGV Platoon under Cyber Attacks

Contributors: Fangjian Li*, Dariusz Mikulski (GVSC), John Wagner*, Andy Dallas (SoarTech), Yue Wang* (PI, * Clemson University)

Unmanned ground vehicles (UGVs) may encounter difficulties accommodating environmental uncertainties and system degradations during harsh conditions. Human experience and onboard intelligence may help mitigate such cases. However, human operators have cognition limits when directly supervising multiple UGVs. Ideally, an automated decision aid can be designed that empowers the human operator to supervise multiple UGVs simultaneously. The human operator needs to trust the UGVs properly in these risky scenarios. In this talk, we consider the scenario where a connected UGV platoon is under cyber-attacks, which may lead to safety disruption and performance degradation. Each UGV generates both internal and external evaluations based on the platoon’s performance metrics. A cloud-based trust-based information management system is designed to collect these evaluations to detect abnormal UGV platoon behaviors. To deal with inaccurate information due to vehicle to cloud (V2C) cyber-attacks, the RoboTrust algorithm is designed to analyze vehicle trustworthiness and eliminate information with low credit. Finally, a human operator scheduling algorithm is proposed for the teaming of the operator and the UGVs when the number of abnormal UGVs exceeds the limit of what the human operators can handle concurrently.

1B1: Computationally Efficient Models for Electro-Magnetic-Structural Dynamics, Heat Convection, and AC Resistance for Electric Machines within Electrified Powertrains

Contributors: Chenyu Yi, Yuanying Wang, Heath Hofmann, Bogdan Epureanu (University of Michigan)

Vehicle electrification involves an increased use of electric machines in vehicle powertrains. The design of these machines, and the corresponding powertrains, requires coupled electromagnetic-structural and electromagnetic-thermal analyses. The coupling of electromagnetic-structural (EMS) phenomena has not been addressed accurately in previous studies. Furthermore, the development of computationally efficient thermal models of electric machines is necessary to obtain accurate temperature information for both powertrain simulation and real-time applications. In this study, a novel hybrid electric vehicle powertrain architecture was designed, modeled, and optimized. A multi-physics reduced-order-model was built to capture EMS coupling with high accuracy and low computational cost. A computationally efficient heat convection model for low/medium rotor speeds has been proposed. Finally, an accurate and computationally efficient AC resistance model has been developed to accurately capture conduction loss for the thermal model.

1B2: Advanced Battery Diagnostics: Decode the information in Electrode Swelling

Contributors: Anna Stefanopoulou (PI), Jason Siegel (Co-PI), Peyman Mohtat (student) (University of Michigan); Yi Ding, Matt Castanier (GVSC); Aaron Knobloch (GE); Dyche Anderson (Ford)

The battery state of health (SOH) is currently estimated by determining capacity (cyclable energy) and cell resistance (power capability). These parameters can save your vehicle or your robot from getting stranded with empty or weakened batteries so they are critical. Unfortunately, estimating these parameters with high confidence can only be done under certain discharge patterns. This uncertainty is masked by overly conservative estimations of the vehicle range, especially under cold temperatures. Knowing the true state of the battery such as the Loss of Active Material (LAM) in the anode and cathode along with the Loss of Lithium Inventory (LLI) allows us to push the cells to their true limit instead of a conservative terminal voltage limit, without inducing more degradation or hazardous shorts from lithium plating.

In this presentation, we’ll take you from the hypothesis of identifying the electrode-specific SOH parameters by observing the cell expansion via a force measurement (year 1), to the intrinsic shift of the phase transition of the battery material as it ages (year 2). This shift creates an “aging signature” like a wrinkle that we can observe in the measured force more clearly than in the measured voltage, especially at relevant discharge ranges and rates. In year three, we will collect data from aged cells and at higher discharge rates to further address the critical real-world considerations and document the evolution of the electrode “age wrinkles” in the electrical and mechanical domains.

1B3: AVPTA Optimization of Scalable Military Fuel Cell Hybrid Vehicles

Contributors: : Jason Siegel, Youngki Kim, Niket Prakash, Miriam Figueroa, Hadi Abbas (University of Michigan); Denise Rizzo, Ben Paczkowski, Andrew Wiegand (HVSC); Buz McCain (Ballard)

This research projects highlights a tech transfer effort to develops scalable physics-based modeling and simulation tools to address right-sizing and power split control for Proton Exchange Membrane (PEM) Fuel Cells (FC) hybridized with a lithium-ion battery pack to meet the high power demand requirements of military drive cycles and extreme environmental conditions in temperature and humidity. The models need to scale for vehicle powertrains ranging from 300W robots to 250kW vehicles to a 750kW armored tank.

When hybridizing a power train, it is challenging to size the energy buffer (EB) or lithium-ion battery and range extender (PEMFC) because the drive cycle, the control policy, and the hardware architecture all affect the optimal size and cooling requirements. The PEMFC system models include dynamics and parasitic losses associated with balance of plant components such as hydrogen storage, compressor, pumps, fans and radiators, that capture the relevant tradeoffs in designing hybrid systems. The key performance metrics include power output, fuel consumption, startup time, volume and weight. The developed control strategies are compared with the baseline optimal solution found using Dynamic Programming. In the second year of this project we also investigated the optimization of the vehicle velocity trajectory, assuming autonomous operation, which could yield up to 15% reduction in energy consumption.

1B4: Computational Discovery of Hydration Reactions for Thermal Energy Storage

Contributors: Steven Kiyabu, Donald J. Siegel (PI, University of Michigan); In-Ho Lee, Denise Rizzo (GVSC); Mike Veenstra (Ford)

Salt hydrates are promising materials for thermal energy storage (TES) due to their high energy densities, moderate cost, and potential for reversible operation at temperatures relevant for ground vehicle applications. While a growing number of salt hydrates have been investigated in the literature for their use in TES, many potential compositions have yet to be explored. The goal of this work is to identify new salt hydrates that can outperform known materials. This will be accomplished using a combination of high-throughput computational screening and machine learning (ML). Several machine learning algorithms were trained on an existing database of known salt hydrates to predict the enthalpy for a given TES reaction. To identify the best input features, thousands of combinations of ionic and structural properties were generated. The most information-rich features were identified via a genetic algorithm. The predictive power of the resulting ML models was assessed on a database of 1,824 hypothetical hydrates. Density Functional Theory calculations were also performed on these hypothetical hydrates to more accurately determine the enthalpy of reaction. Based on these calculations, several promising materials were identified. Revealing the chemical features that are most relevant to TES performance allows for the development of predictive models that can accelerate the discovery of new materials.

1C1: Physics‐Based Multiscale Continuum‐Discrete Deformable Terrain Model for Off‐Road Mobility Simulation

Contributors: Hiroyuki Sugiyama (PI), Hiroki Yamashita, Guanchu Chen (University of Iowa); Paramsothy Jayakumar, Yeefeng Ruan (U.S. Army CCDC Ground Vehicle Systems Center); Mustafa Alsaleh (Caterpillar)

Off-road mobility is the key factor for the U.S. Army to achieve missions and to ensure survivability of soldiers. Physics-based simulations, therefore, play a crucial role in accurate mobility prediction as well as reliable operational planning. Although many deformable soil models have been proposed using either finite element (FE) or discrete element (DE) approaches, phenomenological constitutive assumptions in FE models could lead to inaccurate prediction of complex granular terrain behavior and DE soil models are computationally intensive, especially when considering a wide range of terrain. To eliminate the limitations of existing deformable terrain models, we propose a new hierarchical FE-DE multiscale tire-soil interaction simulation capability that is fully integrated into a monolithic flexible multibody vehicle dynamics solver. To leverage high-performance computing, a scalable multiscale computational algorithm is developed by hybrid shared- and distributed-memory parallel computing scheme, and substantial reduction of computational time is demonstrated as compared to existing pure DE models. Validation results of the proposed multiscale off-road mobility solver are also presented for triaxial soil test and soil bin mobility test conditions. A scalable co-simulation algorithm to enable parallelized full vehicle simulation is also presented. Finally, to ensure the confidence of the proposed model for the NATO benchmark ground vehicle problems, ongoing validation effort using the NextGen-NATO Reference Mobility Model (NG-NRMM) Cooperative Demonstration of Technology (CDT) test data is outlined.

1C2: Massively Parallel Solvers for Complementarity Problems

Contributors: Saibal De, Eduardo Corona, Shravan Veerapaneni (PI) (University of Michigan); David Gorsich, Jayakumar Paramsothy (U.S. Army GVSC)

Enforcing contact constraints accurately in many-body simulations is extremely challenging yet critically important for achieving high fidelity. Optimization-based discrete element methods enforce contacts geometrically through complementarity constraints leading to a differential variational inequality problem. Compared to force penalty methods, they allow for the use of much larger time steps at the expense of solving a nonlinear complementarity problem each time step. We present our recent work on accelerating the solution of complementarity problems using a novel Tensor Train decomposition approach. We demonstrate that this approach displays sublinear scaling of precomputation costs, may be efficiently updated across Newton iterations as well as across simulation time steps, and leads to a fast, optimal complexity solution of the Newton step. This allows our method to gain an order of magnitude speedups over state-of-the-art preconditioning techniques for moderate to large-scale systems, hence mitigating the computational bottleneck of second order methods. Next, we discuss our MPI implementation of first and second order solvers for complementarity problems and demonstrate scalability on thousands of processors.

1C3: Localization, mapping, and path planning performance from LIDAR point-clouds

Contributors: Sam Kysar, Akhil Kurup, Joseph Rice, Alex Gall, Derek Chopp, Parker Young, Jeremy Bos (PI, Michigan Tech)

We compare the performance of two LiDAR systems in mapping, localization, and path-planning in off-road unstructured environments. Both LiDAR systems produce 3D point clouds by spinning a sensor head featuring a discrete number of vertical laser channels. The two systems differ in the number of vertical channels and the scan rate or angular resolution. Specifically, one sensor has more vertical channels while the other has a higher angular resolution; the number of points produced per rotation is similar. Performance is compared by comparing filtered-maps produced by the two systems and the localization accuracy available via iterative closest point matching. Path planning performance is evaluated via the number of valid paths available from random starting and goal positions and overall path length. Comparisons are drawn from a detailed simulation model and initial data from outdoor testing campaigns will be presented.

1C4: Terrain Strength Characterization Using Remote Sensing

Contributors: Jordan Ewing, Thomas Oommen (Michigan Technological University); Jayakumar Paramsothy (GVSC); Russel Alger (Keweenaw Research Center)

Determining the terrain strength characteristics is critical for achieving accurate mobility performance prediction as well as reliable operational planning using the Next Generation NATO Reference Mobility Model (NG-NRMM). Present methods to derive terrain strength for the NG-NRMM rely heavily on in-situ soil strength measurements, which is difficult to obtain from unknown territories and combat zones. In this study, we demonstrate the application of thermal and hyperspectral remote sensing to characterize the terrain strength. The results demonstrate that the apparent thermal inertia is correlated with the stiffness of the soil measured using a geogauge. The hyperspectral imaging shows promise in identifying the soil types, as well as, the approximate water content within each of the soils. Preliminary results indicate that thermal and hyperspectral imaging information together can be valuable for terrain strength characterization using remote sensing.

2A1: The Role of Roles: Adaptive Haptic Shared Control for Remote Steering of an Unmanned Ground Vehicle

Contributors: Brent Gillespie (PI), Akshay Bhardwaj, Xun Fu (University of Michigan); Paramsothy Jayakumar (GVSC); John Walsh (Ford)

Smooth transitions of authority between a human driver and automation system are critical to safe driving. Haptic Shared Control (HSC) offers a means to combine human and automation control over extended time periods or during a transition. Authority in HSC is determined by the relative impedance of the two agents. Previous results demonstrated the promise of HSC for improving team performance and highlighted the role of haptic communication. An automation system that adapts its impedance as a function of estimated driver impedance, alertness, capability, or intent and road conditions might further enhance team performance. We have built a steering wheel that senses grip force and have determined that grip force varies systematically with driver impedance such that grip force serves as a suitable proxy for impedance. We have prototyped an adaptive automation system that decreases its impedance with increasing grip force and are conducting driving simulator experiments to compare team performance under adaptive HSC schemes to team performance in conventional HSC and other schemes.

2A2: Mutually-Adaptive Shared Control between Human Operators and Autonomy in Ground Vehicles

Contributors: Ruikun Luo, Yifan Weng, Jessie Yang, Jeffrey Stein, Matt Reed, Victor Paul, Mark Brudnak, Paramsothy Jayakumar, Vishnu Desaraju, Tulga Ersal

Successful shared control between a human operator and autonomy in ground vehicles critically relies on a mutual understanding and adaptation. Various shared control schemes have been presented in the literature for arbitrating the control authority between the human operator and autonomy. However, most of the existing schemes are not adaptive to the human’s needs and capabilities, and the few adaptive ones do not consider an important human factor, namely, mental workload.

This collaborative project is based on the hypothesis that estimating the mental workload of the human operator in real-time and adapting the shared control scheme accordingly could lead to a more seamless and successful shared control. This presentation will summarize our progress to date to explore this hypothesis. In particular, the presentation will comprise two parts. In the first part we will focus on the challenge of estimating mental workload in real-time and present a novel Bayesian inference based approach that leverages two computational models to estimate human’s workload from two different physiological measurements — gaze trajectory and pupil size. A pilot user study will be reported that shows that the proposed method can estimate human’s workload with only a 4 second-time window and achieve 70% accuracy. In the second part we will present the design of an adaptive shared control scheme. Preliminary human subject studies show that the adaptive scheme can achieve similar vehicle performance as the non-adaptive one with less control effort for the human in a controlled moderate mental workload condition.

2A3: Modeling Bi-directional Trust in Semi-autonomy for Improved System

Contributors: Dawn Tilbury (PI), Lionel Robert, Xi Jessie Yang, Huajing Zhao, Connor Esterwood, Hebert Azevedo Sa (University of Michigan); Victor Paul, Ben Haynes (U.S. Army GVSC); Mitch Rohde (Quantum Signal, LLC)

Automated Vehicles can help drivers to improve the safety and efficiency of transportation, as well as to conclude several activities simultaneously. Trust is a key factor for drivers to leverage vehicle’s autonomy, by sharing the control with automated driving systems (ADS). The use of ADS, however, can result in risky and uncertain situations, as smart systems are not perfectly reliable. In this scenario, it is essential that designers understand the relationship between the risks involved, drivers’ trust attitudes to automation, and the joint performance in cooperative driving with a non-driving task involved. The objective of this project is to establish a methodology to decide in which conditions the driver is likely to assume or to relinquish control authority to the vehicle, and vice versa. We intend to develop computational models to represent humans’ trusting behaviors, especially when operators should respond to systems’ imperfections, and to optimize driving and non-driving task performances based on the trust and different risk conditions. In this presentation, we will discuss the results obtained in subject experiments being conducted and show the details of the preliminary models resultant from analyses concluded.

2A4: Physics Enhanced AI: A Novel Hierarchical Lane Detection

Contributors: Huei Peng (PI), Pingping Lu, Yiqun Dong, Minghan Zhu (University of Michigan); David Gorsich, Christian Balas, Andrea Simon (GVSC); Eric Tseng (Ford)

Artificial Intelligence (AI) has been in active development for years. However, the “black-box” nature makes it hard to explain and difficult to guarantee robust performance. In this project, we study the fundamentals of complementary use of domain knowledge and DNNs to understand when the traditional knowledge is useful & how they can be used together with AI. We focus on lane detection as the target application, a critical function for automated driving. A novel hierarchical detection architecture is proposed which attempts to solve the lane detection problem starting from holistic scene understanding, and then proceeds to the lane line detection problem. This is different from the one large/deep NN approach that many adopted in the literature. The two main contributions of our research are: (1) a hierarchical semantic segmentation network for scene feature extraction, and (2) a rule-based multi-lane parameter optimization method as the lane inferring module. We applied the proposed design process by using data from Cityscapes, Vistas and Apollo, and evaluate the performance on several other datasets: Tusimple, Caltech, and KITTI, as well as Mcity-3000. The proposed approach outperforms the state-of-art lane detection models such as SCNN and LaneNet, and demonstrated better robustness. A real-world experiment is also conducted on the Mcity test vehicle, which shows preliminary but promising lane and curvature detection results compared with the Mobileye.

2B1: A Hybrid Thermal Bus Cooling System for Military Ground Vehicle and Electric Motors

Contributors: John Wagner, Richard Miller, Shervin Shoai Naini, Junkui (Allen) Huang, (Clemson University); Denise Rizzo, Katie Sebeck, Scott Shurin (U.S. Army); Arun Muley, David Blanding (Boeing Research & Technology)

Hybrid electric ground vehicle performance, efficiency, and reliability improvements may be achieved using an enhanced thermal management system featuring passive and smart active cooling solutions. This innovative cooling system architecture includes a high thermal conductivity “cradle” structure for electric motors and/or combustion engine interfaces which offer maximized heat transfer surface area for conduction and convection cooling. The collected heat then travels through a flexible “thermal bus” to one or more heat exchangers. This thermal bus design offers high thermal conductivity materials, carbon fibers, and two-phase passive devices as well as a computer-controlled liquid cooling system for reduced energy consumption and noise generation. A suite of mathematical models and computational simulations describe the hybrid cooling system’s thermal behavior with respect to various driving profiles and operating conditions. The numerical findings are validated through bench top experimental testing with electric motor emulation and various thermal management system configurations. The project results demonstrate that the passive cooling strategies can offer adequate cooling in low to moderate heat dissipation demands often occurring during silent sentry operation while excess heat is removed through the conventional liquid cooling system. Preliminary results show that up to 33% power reduction over convection cooling can be achieved with the thermal prototype subject to various loads.

2B2: Boundary Conditions for Predictive Combustion Simulation

Contributors: Marcis Jansons (PI), Amir Farhat, Taewon Kim (Wayne State University); Peter Schihl (GVSC); Bruce Geist (FCA)

Boundary conditions significantly impact the combustion processes occurring in reciprocating piston engines used in most military vehicles. Injector and wall temperatures, through fuel spray and heat flux rates, affect ignition delay, combustion phasing and duration, and emissions. Values of boundary conditions representative of real combustion systems are thus essential to high fidelity, predictive combustion simulations. In direct-injection compression-ignition systems, fuel and fuel vapor distribution is influenced by the liquid fuel temperature and the physical properties of the fuel used. Optical diagnostics including infrared imaging are applied to characterize in-cylinder and injector tip temperatures to determine their effect on alternative military fuel sprays and subsequent combustion behavior.

2B3: Ignition Studies for Kinetic Mechanism Development and Validation

Contributors: Shuqi Cheng, André Boehman, Margaret Wooldridge, Marcis Jansons, Angela Violi, Peter Schihl, James Anderson, J. Timothy Edwards

The research objectives are to generate ignition data for the purpose of quantifying fuel reactivity and providing high fidelity data for developing reaction theory and reaction mechanisms, and for developing and validating reduced kinetic mechanisms for use in engine simulations. To achieve these objectives, an experimental approach relies on an existing research platform, a modified CFR Octane Rating Engine (used in a previous ARC 4.17 project) with added instrumentation to generate the data needed for the validation of kinetic mechanisms and optimization of the reduced mechanisms. In this year of the project, we will use a newly commissioned GC-MS/FID instrument to determine and quantify stable intermediate species emitted by the CFR engine, thereby determining reaction pathways which serve as a target for numerical simulations to match.

2C1: Investigation and Optimization a Structure to Provide Energy Loss built from an Elastic Material

Contributors: Andrew montalbano, Gang Li, Nicole Coutris, Georges Fadel (PI, Clemson University)

Hyperelastic materials such as rubber provide hysteretic energy loss when loaded. In contrast, elastic materials such as steel or titanium do not. Previous works show that designing a structure consisting of cells of curved-bistable beams produces a structure that provides energy loss. This research focuses on exploring the mechanisms behind energy loss as to maximize the energy lost by the structure. Both a single cell and several meshes of these cells are investigated. Ultimately, this research produces a surrogate model of the energy lost by these cells, discusses system behavior, and optimizes the energy lost by a single cell and various uniform and non-uniform meshes made from elastic materials.

2C2: Fatigue Resistance Optimization of Armored Vehicle Structures

Contributors: Daniel Sinnott (presenter), Carly Mayhood, Pingsha Dong, Nick Vlahopoulos (University of Michigan); Martin M. McDonnell III, Matthew J Rogers, Ravi Thyagarajan (US Army); Nam Purush (BAE Systems)

As it is stated in [GVSC 30-Year Strategy, January 2016] “The Army Chief of Staff set his priority as Readiness. Readiness for ground combat is – and will remain – the US Army’s #1 priority…Readiness is #1, and there is no other #1.” Producing vehicles that reliably meet structural design lives reduces the need for repairs and increases their readiness for combat operations. In armored vehicles, the structure’s fatigue life is dominated by the welded joints that typically have various complex geometric configurations with different material combinations to meet the ballistic performance and structural lightweighting requirements. The situation will become even more severe for autonomous vehicles for which the dynamic loads are not limited by the presence of occupants (for example safety and comfort). This research is developing a fundamental understanding of unique fatigue behaviors associated with thick plate joints used in armored vehicle structures through computational modeling and selected laboratory testing. It is establishing a theoretical framework for data transferability of fatigue test data from different joint types, loading modes, thicknesses, and material combinations relevant to applications in armored vehicles. New algorithms that incorporate the new structural fatigue life evaluations for optimizing armored vehicle structures will be developed. Results from experimental and analytical efforts completed during the first year of this project will be presented.

2C3: Data-Driven Reliability of Repairable Systems using and Effective Age Approach with a Limited Failure Population

Contributors: Themistoklis Koutsellis (GSRA, OU), Zissimos P. Mourelatos (PI, OU), Vijitashwa Pandey (co-PI, OU), Monica Majcher (GVSC)

Most engineering systems are repairable. Their components are repaired after a failure for the system to become operational. In this research, a Generalized Renewal Process (GRP) model quantifies the reliability of a repairable system using the concept of effective age. The model considers repair assumptions such as “same-as-old,” “good-as-new,” “better-than-old-but-worse-than-new” and “worse-than-old,” and is suitable for reset and depot maintenance strategies as well as warranty prediction and forecasting of vehicle fleets. In both maintenance and warranty, it is desired to estimate the Expected Number of Failures (ENF) after a censoring time, using collected failure data before the censoring time. We will present data-driven numerical methods to estimate the ENF of a repairable system accounting for the existence of a small subpopulation with a high failure rate. The latter, known as Limited Failure Population (LFP), may dominate early failures. The ENF of a repairable system can be predicted accurately even with a small number of observed failures increasing the usefulness of our approach for many practical reliability problems where the collection of a large amount of data is not possible or economical. All developments will be demonstrated using representative examples including vehicle production data.