Every public bridge in the United States over 6.1 m has been inspected at least every 24 months since the National Bridge Inspection Standards (NBIS) followed the Silver Bridge collapse of 1967; most other countries run comparable programmes on 2- to 6-year cycles. The routine inspection is visual, hands-on where it matters, and produces condition ratings and a list of defects. Special inspections add to it:

• Fracture-critical (NSTM) inspection, hands-on, arm's-length inspection of steel tension members whose failure would collapse the bridge: two-girder systems, truss chords, pin-and-hanger assemblies.
• Underwater inspection, divers or sonar at intervals of up to 60 months, for scour and for the condition of piers below the waterline.
• Non-destructive testing, ultrasonic and magnetic-particle testing of welds, ground-penetrating radar and half-cell potential surveys of decks, acoustic monitoring of tendons.

Drones, lidar and image analysis are making inspection faster and safer; they do not yet replace the inspector's hand on a pin.

The NBI condition rating scores deck, superstructure and substructure on a 0–9 scale; a bridge whose lowest score is 4 or less is poor, formerly "structurally deficient", which means it needs attention, not that it is about to fall. Element-level inspection is finer: each element (a girder, a bearing, a joint) is quantified in four condition states from good to severe, giving owners a database for prioritising money.

1 good2 fair3 poor4 severe

The rating is a management number and a legal record. It drives the inspection interval, the load-rating cycle, and the funding programme; and when a bridge fails, it is the first document anyone reads.

Element data follow the AASHTO Manual for Bridge Element Inspection, which defines national bridge elements (decks, girders, columns, bearings), bridge management elements (joints, wearing surfaces, protective coatings) and elements an agency defines for itself, each with four condition states and standard defect descriptions. Since MAP-21 (2012), US federal law has required element data for bridges on the National Highway System. Other countries grade differently, Korea with safety grades A to E and Japan with judgement categories I to IV (section 10.7), but all of them separate the condition of parts from the verdict on the whole.

A load rating answers one question: what live load can this bridge carry safely today? The AASHTO Manual for Bridge Evaluation (MBE) computes a rating factor for each member and limit state:

Three levels: the design load rating at inventory (γ_LL = 1.75) and operating (1.35) levels with HL-93; the legal load rating with the state's legal trucks, which decides whether the bridge must be posted with a weight limit; and permit ratings for individual overweight vehicles. A posted bridge is a rating factor below 1.0 at the legal level; a closed bridge is one where the rating falls below a small fraction of legal load. The site's rating tool runs the LRFR arithmetic for a girder.

Bridges age in predictable ways, and the inspector learns the pattern by material:

ASR crackingfatigue crackcorrosionscour

concrete

Cracking (shrinkage, flexural, thermal), delamination and spalling from corroding reinforcement, alkali–silica reaction (map cracking and expansion), freeze–thaw scaling, efflorescence marking where water travels.

prestressed

Corrosion of strand at girder ends and under leaking joints; voided ducts; broken wires detected by acoustic monitoring or magnetic flux.

steel

Section loss from corrosion, especially at girder ends below joints; fatigue cracks at connection plates, cover plates and web gaps; frozen bearings; paint failure.

substructure

Scour, settlement, tilting, cracked pier caps at bearing seats, undermined footings, damaged pile jackets in the splash zone.

appurtenances

Leaking joints, blocked drains, damaged barriers, the cheap problems that cause the expensive ones.

Most bridges are repaired, not replaced. The toolkit is large: deck overlays (latex-modified or silica-fume concrete, polyester polymer) and full deck replacement; cathodic protection and electrochemical chloride extraction for corroding reinforcement; external post-tensioning and bonded CFRP strips to add flexural capacity; steel plates and section replacement on corroded girders; bearing and joint replacement (the commonest job of all); pier jacketing; and scour countermeasures.

jacking for bearing replacementexternal post-tensioning

Two rules from experience. First, understand why the defect happened before fixing it, or the fix fails the same way. Second, strengthening changes the load path, adding stiffness attracts load, so a strengthened member must be analysed as part of the new system, not patched in isolation. The economics are set by traffic: a lane closure on an urban bridge costs more per day than the repair.

Repair selection starts from the cause found in the investigation. Chloride-induced corrosion is the commonest case, and the commonest mistake is patching it alone: clean new concrete in a patch turns the contaminated concrete around it into the anode, and corrosion restarts at the patch edge within a few years, the incipient anode effect.

Bonded FRP is quick and light, but the unstrengthened member must still carry a reduced load on its own in case the FRP is lost to fire or impact, which caps the gain (ACI 440.2R sets the limit). External post-tensioning adds capacity, relieves existing stress, and can be inspected and replaced.

Every rule in this chapter exists because a bridge fell. Silver Bridge (1967): a single eyebar cracked, and the NBIS was born. Mianus River (1983): a pin-and-hanger assembly corroded, and fracture-critical inspection followed. Schoharie Creek (1987): scour under a footing, and the scour programme. I-35W (2007): undersized gusset plates that thirty years of inspection could not have caught, and the rule that gusset plates be load-rated. Morandi (2018): corrosion in encased stays that could not be inspected; FIU (2018): a construction-stage failure; Baltimore's Key Bridge (2024): a ship far larger than the pier protection was designed for.

The site's failure library holds 300 cases classified by mechanism. The pattern that repeats: scour, fatigue and fracture in non-redundant steel, vessel and vehicle impact, construction stages, and hidden corrosion. Read the cases in your own bridge's family before you inspect it.

An inspection programme is a schedule of different kinds of looking, each with its own depth and interval. The 2022 revision of the US National Bridge Inspection Standards names eight: the initial inspection, which records the baseline of a new bridge within three months of opening; the routine inspection, the regular full visual examination; the underwater and NSTM inspections for the parts that need divers or arm's-length access; the damage inspection after a vehicle strike, flood, fire or earthquake; the in-depth inspection of one member or area, usually with testing; the special inspection that follows a known defect between routine visits; and the service inspection, a short check that fills the gap when routine visits are far apart.

The default routine interval is still 24 months, but the rule now lets an owner move away from it in two ways. Under Method 1, a bridge with a component rated 3 (serious) or worse is inspected every 12 months, and a bridge in good order (condition ratings of 6 or better, a load rating factor of at least 1.0, a sound scour rating and adequate clearance, among other criteria) may go to 48 months. Under Method 2, the owner runs a documented risk assessment and may set routine and underwater intervals of up to 72 months; a routine interval longer than 48 months must be bridged by a service inspection. Underwater inspection runs on a 60-month default, cut to 24 months where underwater elements are in serious condition, and a new bridge receives its first underwater and NSTM inspections within 12 months of opening.

The logic of risk-based intervals, set out in NCHRP Report 782, is to inspect most where things change fastest and where a surprise would cost most. A post-tensioned box with a history of grout voids, a truss with pin-and-hanger details and a bridge over a sand-bed river deserve more attention than a young precast bridge on dry land, whatever their condition rating today. The assessment scores details known to perform badly, exposure, loading and consequence: detour length, traffic carried, what lies below.

Other countries settle at similar rhythms. The UK's CS 450 calls for a general inspection every two years and a close principal inspection every six. Germany's DIN 1076 sets a major inspection every six years with a simpler one in between. Japan has required close-up visual inspection of every road bridge at five-year intervals since 2014, sorting each into four judgement categories from I (sound) to IV (urgent action). Korea's facilities safety act ties frequency to the safety grade, A to E: a regular safety inspection at least twice a year (three times for grades D and E); a detailed safety inspection every three years for grade A, every two for B and C and every year for D and E; and, for Class 1 facilities such as cable-supported and long-span bridges, a detailed safety diagnosis every six, five or four years.

The rating equation of section 10.3 is short; the work is in its inputs. Take an interior girder of a simple-span bridge built in the 1990s: 30 m span, precast prestressed I-girders at 2.4 m centres under a 200 mm composite deck, five girders, 75 mm of asphalt. The last inspection rated the superstructure 5 (fair): hairline cracks and two small spalls at girder ends, no exposed strand. Check flexure at midspan, Strength I, for the design load HL-93.

The capacity left for live load is the factored resistance less the factored dead load: 0.95 × 1.0 × 9,600 − 1.25 × 3,150 − 1.50 × 450 = 9,120 − 3,938 − 675 = 4,507 kN·m. At the inventory level, RF = 4,507 / (1.75 × 2,610) = 0.99; at the operating level, RF = 4,507 / (1.35 × 2,610) = 1.28.

Reading the result matters more than computing it. An inventory factor of 0.99 does not make the girder unsafe: inventory is the reliability of a new design, and the member sits a fraction below it. The operating factor of 1.28 says it can carry HL-93 at the lower reliability accepted for an existing structure. The design-load rating is a screen, and a member that does not clear it comfortably is rated next for the legal loads: the AASHTO Type 3, 3S2 and 3-3 trucks, the specialized hauling vehicles SU4 to SU7 and the state's own legal trucks, with a live-load factor set by truck volume (MBE Table 6A.4.4.2.3a-1). On a 30 m span the heaviest AASHTO legal truck produces about 2,370 kN·m per lane with impact, under two-thirds of the HL-93 effect, so this girder rates at about 1.5 or more and no posting follows. Had the 32.7 t Type 3S2 rated at RF = 0.80 instead, the MBE posting relation for a truck of weight W, posting load = (W / 0.7)(RF − 0.3), would give 32.7 / 0.7 × 0.50 ≈ 23 t; below RF = 0.3 the bridge becomes a candidate for closure.

Four inputs move the answer more than the arithmetic does. The distribution factor: the code formula is deliberately conservative, and a refined analysis or a diagnostic load test often lowers it. The section as it is: lost strand, steel section loss and a milled deck, measured rather than assumed. The dead load as it is: an overlay laid on the original surface is a common hidden debit, and the MBE allows γ_DW = 1.25 instead of 1.50 when the wearing surface has been measured. And every limit state: shear near the supports, concrete tension at Service III for prestressed girders, permanent deformation at Service II for steel, and the connections, gussets and bearings that I-35W showed can govern. Many older bridges still carry a load factor rating (LFR) on file, with the HS-20 truck and live-load factors of 2.17 (inventory) and 1.30 (operating); bridges designed to LRFD are expected to be rated by LRFR.

Steel bridges are rarely lost to a single overload; they are lost to a crack that grew for years. Fatigue cracks start at welds, holes and abrupt changes of section, where stress concentrates and residual tension from welding is already present, and they grow a little with every truck that changes the stress, the stress range. AASHTO LRFD (Article 6.6.1.2) sorts details into categories A to E′ by how badly they concentrate stress. Plain rolled material, Category A, survives a constant-amplitude stress range of 165 MPa indefinitely; the end of a welded cover plate on a flange thicker than 20 mm, Category E′, only 18 MPa, a ninth as much.

The older steel stock carries two well-known traps. Cover-plate ends on the rolled beams of the 1950s and 60s, then a tidy way to add capacity, cracked within years on heavily trafficked routes; the cracks found at the Yellow Mill Pond bridge in Connecticut in the 1970s made the E′ detail famous. Distortion-induced fatigue in web gaps: until the mid-1980s, cross-frame connection plates were often stopped short of the tension flange to avoid welding to it, leaving a short strip of web that bends every time adjacent girders deflect differently. These cracks grow under stresses no design calculation predicted, and LRFD now requires connection plates to be attached to both flanges (Article 6.6.1.3.1).

When a crack is found, find its tips with dye penetrant or magnetic particle testing, record it on a sketch, establish whether it lies in a primary tension member, and stop it. The standard field repair is a crack-arrest hole drilled to remove the sharp tip, typically 25 mm or more in diameter, with its edge confirmed crack-free; the lasting repair changes the detail: bolting the connection plate to the flange, or freeing the web gap so it can flex without strain concentration; bolted splice plates across a cover-plate end; or peening the weld toe to put it in compression.

Fracture is the sudden end of a crack's growth, and it needs three things together: a crack-like flaw, tensile stress and steel with low toughness at the service temperature. Designers control toughness by specifying Charpy V-notch energy by temperature zone, with stricter values for fracture-critical members, which are also fabricated under the fracture control plan of AASHTO/AWS D1.5. Geometry matters as well. In December 2000, girders of the Hoan Bridge in Milwaukee cracked almost through their full depth, in cold weather and without warning, from a welded detail where a lateral gusset plate, a connection plate and the web met with almost no gap between their welds. That triaxial constraint left the steel unable to yield, and the fracture started with almost no fatigue growth before it.

Since 2022 US rules call these members nonredundant steel tension members (NSTMs) rather than fracture-critical members. Two AASHTO guide specifications of 2018 let an owner show by refined analysis that a two-girder bridge or a truss has enough system redundancy, or that a riveted or bolted built-up member has enough internal redundancy, to survive the fracture of one component; a member so shown is not an NSTM. NSTM inspection is hands-on at arm's length, normally every 24 months: every weld termination, cope, splice and pin is examined, and findings are carried from report to report on sketches so that crack growth can be measured. Pins are tested by ultrasound from their ends, because the cracks and corrosion that matter sit where the hanger plates bear and cannot be seen; many owners have fitted catcher systems behind old pin-and-hanger details since Mianus River.

More bridges are lost to water than to any other cause, and the parts water attacks are the parts nobody sees.

FHWA practice defines three intensities of underwater inspection. Level I is a visual and tactile swim-around of every underwater element, enough to find gross damage and scour. Level II adds cleaning of marine growth from representative areas, on at least a tenth of the elements, so that cracks, section loss and spalls can be measured. Level III is a detailed investigation with NDT, coring or thickness gauging, done where Level II raises a question. In the near-zero visibility of most rivers divers work by touch; multibeam and scanning sonar now give a picture of the bed and pier faces that a diver cannot, and bathymetric surveys repeated over the years show whether a channel is degrading or migrating.

The hardest problem is that scour hides. The deepest scour occurs at the flood peak, and as the flow falls the hole refills with loose sediment, so a low-water inspection probes a level bed resting on a backfilled cavity. Probing rods, sub-bottom profiling and comparison with the as-built foundation level are the defence. A bridge whose foundation depth is simply unknown, which is common for older structures, is a risk in its own right.

A bridge is scour critical when observed or calculated scour (HEC-18) would make a foundation unstable, and a US owner must keep a plan of action (POA) for every such bridge. A good POA is short and operational:

• the monitoring: periodic inspection, flood watch by staff, or fixed instruments such as sonar, float-out devices and tilt sensors on piers;
• the trigger, a river stage, discharge, measured scour depth or tilt, at which the bridge is closed, and the person with authority to close it;
• the detour, and the conditions for reopening, normally a post-flood inspection with diving if needed;
• the permanent countermeasure and its date, so the plan is a bridge to a fix rather than a substitute for one.

Countermeasures (FHWA HEC-23) come in three families: armour (riprap, articulated concrete blocks, grout-filled mattresses, gabions), structural (underpinning, deepened foundations, added piles) and river training (guide banks, spurs, bendway weirs) that moves the attack away from the bridge. For pier riprap HEC-23 sizes the median stone as D₅₀ = 0.692 (K V)² / [2 g (S_s − 1)], with K = 1.5 for a round-nose and 1.7 for a square-nose pier. At V = 3 m/s and S_s = 2.65 a square-nose pier needs D₅₀ ≈ 0.56 m, placed at least three stones thick and extending about two pier widths from each face. Stone that is too small simply leaves; stone without a filter loses the bed through its gaps. Both failures are found after the first big flood.

Scour is a property of the river, not only of the bridge. The Hintze Ribeiro Bridge over the Douro in Portugal collapsed in March 2001 when a pier, its bed lowered by years of sand extraction, was undermined in flood; 59 people died. Dredging, dams and channel works upstream and downstream belong in the inspection record. In marine bridges, the Gulf's among them, the underwater programme is as much about corrosion: pile jackets and the splash zone, cathodic-protection anodes and potentials, marine growth that adds mass and hides defects, and the ship-impact protection that the Key Bridge collapse made a board-level question.

The inspector's eye and hammer find most defects. Non-destructive testing and evaluation (NDT/NDE) are for the ones they cannot reach: corrosion that has not yet cracked the cover, a void in a grouted duct, a crack inside a pin, a broken wire inside a stay. Every method measures a physical property, not the defect itself, so the working rule is to use NDT to map and to target, then confirm with a few cores, openings or samples before anyone acts on the map.

Half-cell potential is the most used and the most misread. ASTM C876, measured against a copper/copper sulfate electrode, reads potentials more positive than −200 mV as a better than 90% probability that no corrosion is occurring, more negative than −350 mV as a better than 90% probability that it is, and the band between as uncertain. Wet, saturated or carbonated concrete shifts these values, so practice now reads the pattern and gradient over a grid rather than single numbers. Paired with chloride profiles, which show whether chloride at bar depth has passed the threshold for corrosion (commonly taken around 0.2–0.4% by mass of cement, with wide scatter), the survey tells an owner how much of a deck is corroding now and how much will be in ten years.

For post-tensioned bridges the hard question is the tendon. Around 2000, corroded and failed tendons in Florida segmental bridges (Niles Channel, Mid-Bay and the Sunshine Skyway piers) showed that grout voids, bleed water and soft grout can leave strand unprotected inside a sound-looking duct. The toolkit that followed combines anchorage inspection, borescopes through small holes drilled at duct high points, impact-echo and radar for voids, magnetic flux on external tendons and acoustic monitoring for wire breaks. For stay cables, vibration-based force measurement (section 10.12) and magnetic flux over the free length are the standard pair; the anchorage, where most stays corrode, still needs an eye and a hand.

Structural health monitoring (SHM) is inspection by instruments, continuously. Its value lies not in the number of sensors but in the question each one answers. Good reasons to monitor: to follow a known defect between inspections (a crack gauge across a fatigue crack, a tiltmeter on a scour-critical pier); to check a design assumption that matters (the wind response of a long deck, joint movements, stay forces); to run the bridge (wind speed for closing it to high-sided vehicles, ship detection, ice); and to measure the real loading, where weigh-in-motion records the trucks actually crossing and feeds fatigue and rating calculations.

Long-span cable bridges now carry monitoring from opening day; Hong Kong's Wind and Structural Health Monitoring System, installed on the Tsing Ma Bridge when it opened in 1997, set the pattern. A typical package combines anemometers, accelerometers on deck, towers and stays, GNSS receivers for global displacement, temperature sensors, strain gauges at fatigue-sensitive details, joint and bearing displacement transducers, weigh-in-motion and embedded corrosion sensors.

The difficulty of SHM is interpretation, not measurement. Temperature dominates almost every signal: a deck's natural frequencies shift between winter and summer by more than moderate damage would shift them. Useful systems therefore remove environmental effects statistically before looking for anomalies; set alarm thresholds in advance, each level (green, amber, red) with an action and a named person who acts; and budget for upkeep, because sensors and loggers last 10–20 years against a bridge's 100. A monitoring system that nobody reads is worse than none: it gives the impression of oversight without the substance.

A digital twin goes one step further: a model of the bridge kept current with what inspection and monitoring report, and used to make decisions. In rising order of ambition it can be an organised asset record (the as-built model with every finding, repair and test located on it); a structural model calibrated to measured behaviour, which can re-rate the bridge the day a defect is found or a heavy permit is requested; or a predictive model that combines deterioration laws with measured loading to forecast when a member will need work. The first is achievable for any major bridge today and pays for itself at the next inspection; the third is still research for most structures. The test of any twin is the question that opened this section: which decision does it improve?

An owner with 2,000 bridges and money for 40 projects a year needs a way to choose. A bridge management system (BMS) does it with five parts: an inventory; inspection data, ideally element quantities in condition states (section 10.2); deterioration models that forecast condition; cost models for each action; and an optimisation that picks the programme giving the best network condition, or the lowest long-run cost, within the budget. AASHTOWare BrM, the successor to Pontis, is the American standard; most countries run a national system of their own.

Deterioration is usually modelled as a Markov chain: each year, each element quantity has a probability of staying in its condition state or dropping to the next, estimated from the history of inspections. The numbers are easy to read. If 95% of a deck's area in state 1 is still in state 1 a year later, the median time in state 1 is ln 0.5 / ln 0.95 ≈ 13.5 years; raise that annual probability to 0.97 with a membrane and regular sealing and the median becomes about 23 years. The model forecasts the spread of an element across states 1 to 4 in any future year without predicting the fate of any single square metre.

Actions fall into three classes, and the economics favour the first. Preservation: cyclical work done regardless of condition (washing salt off steel and bearing shelves, cleaning drains, resealing joints, sealing decks with silane) and condition-based work on elements still in good or fair condition (spot painting, patching, seal replacement). Rehabilitation: deck replacement, strengthening, bearing replacement, full repainting. Replacement. A worst-first policy, spending only on the poorest bridges, looks responsible and slowly ruins a network, because the good bridges deteriorate unattended into the next generation of poor ones. FHWA's Bridge Preservation Guide and most BMS optimisations find that cheap preventive work on fair-to-good bridges buys the most condition per unit of money.

Life-cycle cost analysis compares options on one basis by discounting each future cost C at year t to a present value, PV = C / (1 + r)^t. At a real discount rate of 3%, money spent in 15 years is worth 0.64 of money spent today; in 30 years 0.41; in 45 years 0.26.

The result depends on the discount rate and on service lives that are themselves estimates, so every comparison needs a sensitivity check; whole-life carbon is increasingly reported beside cost.