International Journal of Civil and Structural Engineering Research ISSN 2348-7607 (Online) Vol. 10, Issue 1, pp: (14-20), Month: April 2022 - September 2022, Available at: www.researchpublish.com
Damage Index Combination via Genetic Algorithm: A Highway Bridge Case Study Elizabeth K. Ervin1, Chuangshuo Zeng2 1
University of Mississippi, University, MS, USA 2
Sany Heavy Industry, Shanghai, China
Abstract: Time histories often vary too much to determine root cause of signal shift. Frequency information and extracted modal properties can correlate to structural health through quantitative change metrics. Herein, fiftyone modal-based damage indices are considered in a total of three directions. These metrics are combined via vector resultants and Genetic Algorithm to visualize final relative stiffness results by location of data capture using a color code. This work presents inspection and data mining on an obsolete three-span truss highway bridge from 1953 or 1941 (disputed). Tri-axial deck data was captured in a grid for all three spans, and the new damage detection methodology is applied to Spans 1 and 2 with Span 3 as control. Comparative analysis among the three spans quantified joint effects, and end damages due to both scour and spalling were identifiable. Furthermore, unbiased analyses provided similar results to those biased by visual inspection. A more pointed visual inspection is thus permitted post-analysis. Keywords: Modal Analysis, Data Mining, Genetic Algorithm, Bridge Inspection, Damage Detection.
I. INTRODUCTION This paper applies a new damage detection methodology to locate and quantify damage on a three-span highway bridge. In 2015 the MS Highway 7 Bridge over the Tallahatchie River was being replaced by a single-span prestressed concrete structure by the Mississippi Department of Transportation. The truss bridge was built in either 1953 or 1941 (disputed). Each simple span was “identical” at 120’ long and 26’ wide with 13’-6” vehicle and 20’ overall truss height. The Multi-Function Dynamics Laboratory in the Department of Civil Engineering at the University of Mississippi was given four hours to inspect said bridge before demolition. A quick cursory visual inspection noted (1) Span 1 had piers showing severe scour just above the water line, (2) Span 2 had a widely opened expansion joint at Span 1 showing spalling, and (3) overheight vehicle damage and both ends of the bridge. These items were confirmed by the lead supervisor; in fact, the new bridge opened days early due to his observation of concrete spalling in Span 2. Note that the actual condition of the deck underneath remains unknown. The damage detection methodology compares two states of a structure, usually subsequent inspections of the same structure. Based upon observed bridge condition, Span 3 is considered the “control” in the numerical experiment. That is, Span 1 will be compared to Span 3 to quantify scour effects, and Span 2 will be compared to Span 3 to quantify expansion joint effects. This assumes that data can be affordably and conveniently captured by a bridge inspector; enough sensor time histories are taken to locate and quantify relative weakness; and frequency variations are due to damage rather than operating conditions. On the day of testing, the ambient temperature was steady near 30 oF, and the deck temperature ranged from 33.4oF to 38.8oF. Wind gusts of 15-20mph were predicted; the contractor measured a 13mph steady wind at one point. The X, Y, and Z directions were selected to be the bridge’s longitudinal, transverse, and vertical directions, respectively. Assumed level due to minimal road crowning, each span had 27 test locations: the deck grid was divided into nine points longitudinally at 15’ spacing and three points laterally at 13’ spacing. This results in a total of 81 nodes, but one point on Span 1 was omitted by accident. Each test node was sequentially measured under ambient vibration by a PCB Model 356B18 (1V/g) connected by traditional BNC cabling to a NI 9234 module in a NI CompactDAQ chassis, connected by USB to a laptop running NI LabVIEW (Figure 2). Accelerations were measured at 2048 Samples per second for 30 seconds. Examination of the time histories on Span 1 found a maximum longitudinal X-acceleration of 50.4±1.794mg, a maximum transverse Y-acceleration of 102.7±1.097mg, and a maximum vertical Z-acceleration of 35.8±1.247mg.
Page | 14 Research Publish Journals