Interpreting Impact Echo Data to Predict Condition Rating of Concrete Bridge Decks: A Machine-Learning Approach

Maintaining the structural reliability of highway bridges under a budget constraint necessitates development accurate prediction models bridge deck deterioration to maximize service life while minimizing life-cycle costs. Traditionally, condition is assessed using ordinal discrete indices, referred as ratings (CRs), assigned based on an assessment visible signs deterioration. Nondestructive evaluation (NDE) being increasingly utilized gain objective insights into The impact echo (IE) test common NDE technique that relies acoustic resonance response detect subsurface delamination can lead spalling. However, IE data interpretation largely done manually and connection between results CRs not fully explored. aim this study model spectral characteristics signals quantify integrity decks predict CRs. First, nearest neighbor clustering signal energy distribution in frequency domain conducted generate labels for each (good, fair, poor) automatically. are then input support vector machine (SVM) classification trained tested from Long-Term Bridge Performance (LTBP) set pertaining 38 with recorded collected over span 2 years average. findings indicate proposed capable automatically predicting given raw accuracy 87.5%.