time good Bridge Risk Modeling Current Florida DOT Research condition bad Paul D. Thompson, Consultant
Background Research team Florida State University John Sobanjo (PI) Paul D. Thompson (co-pi) Florida DOT 31,166 structures (bridges, culverts, sign structures, HMLPs, mast arms) 139,669 inspections over >14 years 1,120,031 element inspections 5,033 retired structures
Risks in the Florida inventory Florida is exposed to natural and man-made hazards: hurricanes, tornadoes, flooding, landslides, wildfires, etc. Example: 2004 Hurricane Ivan damaged the I-10 Escambia Bay Bridge in Pensacola Research objective -- develop a framework and implement risk assessment + management models
Natural hazards 4
Natural hazards in Florida Earthquakes are very rare, but... Over half of the hurricane-related damage in the United States occurs in Florida Hurricanes can cause storm surge and waves, deck uplift and transverse forces, impact damage from floating debris, and scour Tornadoes, flooding, and wildfires are also common
Hurricanes Hurricanes classified on Saffir Simpson (1-5) Scale, and related to physical damage observed Spatial Analyses of FEMA Hurricane Maps (HAZUS Software) and FDOT Bridge GIS Layers estimate hurricane wind speeds and return periods at each bridge location Two-parameter Weibull distribution estimated for wind speed at each bridge location Probabilities of each hurricane category are then estimated for each bridge 6
Hurricanes EXAMPLE: A bridge has average annual rate λ for hurricane categories 1, 2, 3, and 4 estimated as 0.05, 0.02, 0.01, and 0.002 The probability of the bridge experiencing a hurricane of category 1 within the next one year is estimated as The probability of the occurrence of hurricane category 1 within the next 10 years is estimated as
Tornadoes In Florida, tornadoes can occur at any time of the year; form on their own, or accompany hurricanes and tropical storms Spatial Analyses of NWS Tornado GIS layer and FDOT bridge GIS layer, identified tornadoes within 1 mi. of each bridge Mean annual occurrence rate estimated for each bridge using Poisson process
Flooding Much of Florida is at or near sea level Failure modes: collapse caused by scour; hydraulic loading on structure; erosion of abutments; and impact from debris Spatial Analyses of FEMA Flood GIS Data, Florida Geographic Data Library (FGDL), and FDOT Bridge GIS layer Identified flood zones and estimated annual occurrence rate for each bridge
Wildfires Hazardous to steel and timber structures Spatial Analyses of Florida Department of Forestry wildfire data (1980 to 2010) and the FDOT Bridge GIS layer Identified wildfire locations within one mile of each bridge Mean annual occurrence rate estimated for each bridge
Man-made hazards 11
Vehicle or Vessel Collision, Fire Conventional vehicular crash, especially involving fuel tankers may cause fire. Vessel/Barge impacts on bridge substructure may also cause fire. Ongoing work on this category.
Advanced deterioration 13
Framing the risk analysis Advanced deterioration may cause unwanted service interruption: Closure or replacement of bridge Replacement of superstructure Load posting Network and bridge level models: Pontis failure probabilities Relate condition to disruption probability, and then to user cost or disutility 14
Data sources FDOT doesn t delete structures, but merely reassigns them to a fictitious District 9 Historical inventory and condition data are retained currently 5033 structures Link to replacement structure is maintained in the userbrdg table Reason for retirement/replacement explained in comments 15
Processing Retired bridges compared with their replacements to identify functional improvements Deteriorated element conditions identified Each bridge manually reviewed and classified for primary reason for retirement 16
Extension of previous research
Markov model estimation Linear regression Traditional method Transition to any worse state Usable models: 172 Min sample: 1500 r 2 : 0.7213 One-step New method Transition to just nextworse state Usable models: 253 (Out of 755 models at the element/environment level) Min sample: 500 r 2 : 0.7217 One-step method makes better use of data without sacrificing explanatory power.
Expert judgment unreliable Ratio of new transition times to old (2000) expert judgment models By element category By element material Joints 3.2 Unpainted steel 1.8 Railing 1.6 Painted steel 1.9 Superstructure 1.7 Prestressed concrete 1.7 Bearings 2.2 Reinforced concrete 2.1 Substructure 2.0 Timber 1.8 Movable bridge equip 1.8 Other material 2.1 Channel 1.4 Decks 1.9 Other elements 1.4 Slabs 3.3 Expert panel under-estimated transition times by a factor of 1.97 on average.
Better model of young bridges Shaping parameter (beta) slows the onset of deterioration
Lessons: Florida and Virginia Success factors for condition modeling: Inspections should consistently record (as condition state data) severe maintenance-related defects as well as safety and function defects Need a reliable way to identify past actions: maintenance, repair, rehabilitation, improvement, and replacement Need to control for relatively new materials (e.g. weathering steel and prestressed concrete) 21
Better model of old bridges Experimentation with several ways of framing the analysis and modeling risk Best model so far: Worst condition state and second-worst Deck, super, and sub elements (< #300) Risk proportional to time in deteriorated states % in worst states compounds over time, based on multiplication of transition probabilities % in worst states -> disruption probability time condition good bad 22
Lognormal risk model Appropriate when explanatory variable is built up by multiplication Based on log of weighted percent in worst and 2 nd - worst states for each inspection For each inspection indicate if bridge underwent retirement, replacement, reconstruction, or posting before next inspection Compute lognormal hazard function and element weights using maximum likelihood estimation 23
Example model Reinforced concrete bridges Decay index: Weighted condition similar to health index, but emphasizes the worst and 2 nd -worst states. 100=worst 24
Project Level Analysis Tool 25
Thank you! Paul D. Thompson www.pdth.com