No abstract

This paper presents a new approach to a robust Gaussian process (GP) regression, creating a non-parametric Bayesian regression estimate robust to outliers. Most existing approaches replace an outlier-prone Gaussian likelihood with a non-Gaussian likelihood induced from a heavy tail distribution, such as the Laplace distribution and Student-t distribution. However, the use of a non-Gaussian likelihood would incur the need for a computationally expensive Bayesian approximate computation in the posterior inferences. The proposed approach models an outlier as a noisy and biased observation of an unknown regression function, and accordingly, the likelihood contains bias terms to explain the degree of deviations from the regression function. We introduce two bias models that handle the bias terms differently, treating a bias as an unknown and fixed quantity or treating a bias as a random quantity. We entail how the biases can be estimated accurately with other hyperparameters by a regularized maximum likelihood estimation. Conditioned on the bias estimates, the robust GP regression can be reduced to a standard GP regression problem with analytical forms of the predictive mean and variance estimates. Therefore, the proposed approach is simple and very computationally attractive. It also gives a very robust and accurate GP estimate for many tested scenarios. For the numerical evaluation, we perform a comprehensive simulation study to evaluate the proposed approach with the comparison to the existing robust GP approaches under various simulated scenarios of different outlier proportions and different noise levels. The approach is applied to data from two measurement

Performance degradation of urethane and epoxy coatings on high performance aircraft due to chemical reversion, while not visible on the surface, can be the result of changes in physical and chemical composition, leading to premature coating failure. Standard lifetime predictions fail to estimate the functional lifetime of urethane coatings on operational aircraft due to the wide variety of environmental factors and combinations of factors that trigger degradation at different rates. To better understand the role that environmental conditions play in the overall lifetime performance of these coatings, a non-destructive technique is necessary to quantitatively assess the degree of degradation/reversion that can occur. The scanning Kelvin probe (SKP) has demonstrated itself to be sensitive enough to detect changes in the alkyl chain lengths of polymers and their terminal groups as well as determine the interfacial diffusion of water through epoxy coatings as a function of the chemical structure and functional groups present within the coating. It has also demonstrated its capability to detect corrosion under coatings at the coating-metal interface. These capabilities make the use of the scanning Kelvin probe technique a compelling approach to detecting and characterizing the degradation behavior of multi-layer urethane and epoxy coating systems as well as any corrosion occurring at the coating-metal substrate interface. The objective of this effort is to develop a non-destructive evaluation technique for measurement and analysis of a multi-layer polyurethane/epoxy coating system subjected to elevated temperature and relative humidity (100 C, 100%RH) degradation conditions. Measurements of changes in the work function of non-exposed and exposed rain erosion coating (REC) system for various times at the elevated conditions have been made to correlate the degree of degradation within the coating system as determined by Raman/FTIR spectroscopy.

Replicate coupons of bare aluminum alloys AA2024-T3, AA6061-T6, and AA7075-T6 are being exposed at two coastal atmospheric test sites, Kennedy Space Center (KSC) and the US Naval Research Laboratory in Key West, FL (NRL-KW). Various analyses including mass loss/mass gain determinations, elemental composition analysis (via scanning electron microscopy-energy dispersive spectroscopy (SEM-EDS)), as well as pit density and pit volume measurements (via 3-dimensional laser/optical microscopy) are being performed to determine if there is a correlation between corrosion rate, corrosion morphology, and surface chemistries of the coupons exposed at the outdoor sites. The sample coupons are being retrieved at 3-month intervals. The aluminum alloy coupons are cleaned until a consistent change in mass between cleaning cycles has been reached per ASTM G1 (Table A1). At this point of the investigation, a cumulative exposure time of 6 months will be discussed and presented. Mass loss data for the bare metal substrates at the two field locations indicate that different mass losses occur at the two locations for similar exposure lengths for all three alloys, but at overlapping time intervals. For the AA7075-T6 alloy, mass gain (negative mass loss values) was measured for the 1Q (3 month) and 2Q (6 month) exposure at the NRL-KW site, whereas mass loss values (positive mass loss values) were determined for the KSC site for similar time exposures; however, replacement replicate AA7075-T6 coupon exposures at the NRL-KW site for the same length of exposure but at a different time interval, exhibited mass loss values (Figure 1). While the observed corrosion rates were different, it was determined that the morphology and elemental composition of the alloy substrates were different as well. Due to the extreme localized pitting attack on the 6061 and 7075 alloys, optical analyses of the coupons indicate that the resulting variability of the mass determinations may be due to incomplete removal of corrosion product from within the subsurface horizontal exfoliating pits populating the exposed coupons. Indeed, the populating of the coupons with these occluded pits was observed to increase with field exposure time at both field exposure sites, resulting in increased variability in mass loss values. These results reinforce the importance of deploying more than a single sample coupon of these aluminum alloys for corrosion rate calculations using mass loss determinations. Pit morphology and elemental analysis of the exposed alloy coupons as a function of exposure time and location will also be presented and discussed. These findings are important in developing accelerated test protocols. In previous laboratory exposure tests it was found that similar mass losses were observed at an accelerated rate in the modified chamber for these alloys versus the field exposures, but the surface morphology of the substrates exposed in the chamber were markedly different from the substrates in the field. Therefore, it may be more informative to consider the elemental composition, pit density and pit morphology of the corroded surface in addition to the mass loss, rather than rely on mass loss alone. Figure 1. Plot of mass loss comparison for aluminum alloys coupons exposed at KSC and NRL-KW for 3 and 6 months. Figure 1

How can we design short-term laboratory corrosion tests to adequately simulate long-term outdoor atmospheric corrosion? To achieve this objective, models need to be devised to rank and quantify critical environmental variables that influence atmospheric corrosion. This effort provides a data-driven analysis of environmental factors that can be used to quantitatively model atmospheric corrosion aluminum alloy 2024-T3 (AA2024-T3) samples placed at three unique coastal sites in the state of Florida, USA, for various time periods over an 18-month total period of exposure. Atmospheric corrosion proceeds via several processes that take place in sequence and/or in parallel across multiple classes of matter: the atmosphere, condensed aqueous solution, organic coatings, oxide scales, precipitated salts, and microstructurally heterogeneous metal alloys. Several physical and chemical phenomena contribute to the process of corrosion, including mass-transport, electrochemical effects, metal dissolution, grain-boundary transport, etc. For this reason it is difficult to directly predict, using fundamental physics or chemical principles, the corrosion rate of a metal in its environment. Likewise, it is difficult to directly extrapolate the results of short-term tests to long-term tests solely from physical principles. A modeling approach that pairs data analytics with scientific insight is required. To support this objective, data was collected for AA2024-T3 coupons based on exposure over an 18-month period at three coastal sites in Florida, USA. Methods to summarize the environmental exposure metrics for each time period were developed using standard statistical metrics (mean, standard deviation) as well as a more complete set of metrics, known as the Catch-22 algorithms, developed by The University of Oxford. The corrosion coupon data was aggregated into a data framework that included the metrics of mass loss per unit area, linear corrosion rate, and a parabolic corrosion constant, as well as generation of additional data by sample differencing that considers cumulative corrosion occurring between time periods. An automated approach was then developed that queries public and/or pay-to-access websites for environmental data to build an exposure profile for a sample placed at a known location (specified by latitude/longitude) and over a given range of dates. We evaluated 288 total variables in the exposure profile. Of these 288 variables, five key variables were determined to have a quantitative effect on the corrosion rate and mass loss per unit area, and these include mean precipitation, the range of temperatures, the minimum wind speed, the standard deviation of ozone exposure, and the maximum solar irradiance. This work serves to help with the selection of appropriate variables to be used in designs for a laboratory-simulation exposure chamber that would mimic service environment and accelerate the development of advanced materials degradation test protocols. The approach developed herein can be applied to other materials of interest, different locations, and adapted to other metrics of corrosion such as localized corrosion depth and volume, due to pitting.