Neutral salt spray tests were conducted on assemblies comprising H59 brass and either electrogalvanized or hot-dip galvanized bolts. The polarization curves, electrochemical impedance spectroscopy (EIS), corrosion morphology, elemental distribution, and corrosion product composition of the H59 brass were systematically characterized. The results demonstrated that upon coupling with galvanized bolts, the formation of a protective Cu₂O film on the H59 brass is significantly weakened, leading to accelerated corrosion. After coupling with electrogalvanized bolts, the i_corr reached a maximum value of 0.21 mA/cm². A corrosion layer predominantly composed of ZnO formed on the sample surface with a thickness of approximately 13 μm, and no penetration or enrichment of Cl was observed in the matrix. More seriously, when the brass was assembled with hot-dip galvanized bolts, the i_corr never dropped below 0.2 mA/cm². A porous and complex Zn-Cu-O-Cl mixed corrosion layer developed on its surface. This loose structure allows Cl⁻ to reach a depth of 55 μm into the matrix and continue causing corrosion. The mechanisms underlying the different corrosion behaviors of H59 brass caused by different galvanizing bolt processes require further investigation.

Using electric vehicles (EVs) for grid support requires reliable flexibility estimates and effective voltage control from limited online information. We propose a behavior-informed interacting multiple-model estimator (BIMM) and an availability-aware voltage-control strategy. The estimator combines predicted behavior-mode probabilities with station-level measurements to estimate aggregate upward and downward energy flexibility and its uncertainty. The controller combines behavior-derived availability priors with online measurements to update station-level reactive-power limits and allocate voltage support. In paired simulations, explicit behavior-prior pooling reduces the normalized continuous ranked probability score (nCRPS) for energy by 54.0% and 61.8% in IID and OOD, respectively, relative to the same-bank no-pooling reference. Relative to the standard interacting multiple-model estimator (Standard-IMM), the normalized root-mean-square error (NRMSE) for energy decreases by 23.36% and 39.48% in the batch-charging scenarios S1 and S2, respectively; energy losses in ordinary IID/OOD operation are higher than those of Standard-IMM. In ordinary control scenarios, mean action-time request-exceedance rates are lower than for fast-measurement-only voltage control (FastOnly). These results demonstrate the use of aggregate behavior information for probabilistic flexibility estimation and availability-aware reactive-power dispatch.