Trained bridge inspectors who could touch the steel found only 65% of known fatigue cracks on painted girders, and from the ground they found 4% (Purdue JTRP, Probability of Detection Study for Visual Inspection of Steel Bridges 2019). That gap is the whole case for acoustic emission, the technique that listens for the stress waves a steel member releases when a crack actually grows. The need has not gone away: the August 2026 National Bridge Inventory table still lists 41,148 bridges in poor condition (FHWA, Bridge Condition by Highway System 2026), and motorists cross poor bridges 163 million times a day (ARTBA Bridge Report 2025). The 2022 Fern Hollow collapse showed the cost of missing a fracture-critical member, after repair recommendations sat in inspection reports from 2005 to 2021 (NTSB, Fern Hollow Board Meeting Summary 2024). We aggregated data from the National Bridge Inventory, the NTSB, Purdue’s S-BRITE center, ASCE, ARTBA, and peer-reviewed acoustic emission field and lab studies. For a broader view of machines that listen for faults, see our acoustic anomaly detection statistics.
TL;DR
- 41,148 of 624,939 US bridges are rated poor in the August 2026 NBI table, about 6.6% by our calculation (FHWA, Bridge Condition by Highway System 2026).
- Motorists cross poor-condition bridges 163 million times a day (ARTBA Bridge Report 2025).
- About 45% of US bridges have exceeded their planned 50-year design life (ASCE Infrastructure Report Card 2025).
- Nonredundant steel tension members must be inspected hands-on at least every 24 months, or every 12 months once rated poor (23 CFR 650.311).
- Fern Hollow’s 447-foot bridge fell about 100 feet after a fracture-critical tie plate failed (NTSB, Fern Hollow Board Meeting Summary 2024).
- AE monitoring detected a propagating fatigue crack on 2 of 3 Minnesota field bridges and an extinguished crack on the third (MnDOT, MN/RC-1999-36, 1999).
- A 48-month NSTM interval is allowed only for bridges built after 1978 under a fracture control plan with no history of fatigue cracks (23 CFR 650.311).
- Fatigue cracks through a test bridge’s girder flanges did not change its global eigenfrequencies, so vibration-based monitoring missed them (EWSHM 2024, Lachinger et al.).
- On a full-size steel railway bridge, AE activity at the crack rose from 0.4 to 9.7 located events per cm per 1,000 cycles (TUV Austria, EWGAE 2024).
- Denoised AE signals were classified with 92.56% (damaged) and 94.27% (undamaged) test accuracy on a bridge-deck steel specimen (PLoS ONE, Li et al. 2025).
- Ambient bridge noise never met all AE fracture criteria in a set during monitoring of a fracture-critical tied arch (UMN CTS, Protocols and Criteria for AE Monitoring 2015).
1. The Inventory: How Many US Bridges Rate Poor in 2026
Every bridge in the National Bridge Inventory (NBI, the federal database of public road bridges) gets condition ratings for its deck, superstructure (the beams and girders) and substructure on a scale from 0 (failed) to 9 (excellent), and a rating of 4 is considered poor (ARTBA Bridge Report 2025). A bridge is counted as poor when any one of those key elements is rated 4 or lower.
The headline count barely moves year to year: 41,148 bridges are rated poor in the August 2026 table, against 41,677 in the 2025 data analyzed by ARTBA. The real story is the backlog behind them: more than a third of the inventory needs work, and every one of those structures still depends on a human inspector reaching the right detail at the right time.
| Metric | Value | Source |
|---|---|---|
| Bridges rated poor vs all bridges, and poor deck area vs total deck area, table dated August 4, 2026 | 41,148 of 624,939 bridges (6.6% by our calculation); 217,000,522 of 4,548,152,281 sq ft (4.8% by our calculation) | FHWA, Bridge Condition by Highway System 2026 |
| Bridges rated poor (structurally deficient), 2025 data | 41,677, down from 42,067 in 2024 | ARTBA Bridge Report 2025 |
| Daily crossings of poor-condition bridges | 163 million | ARTBA Bridge Report 2025 |
| Bridges needing major repair or replacement, 2025 data | 35% of all US bridges: 220,295 spans, including 74,472 that should be replaced | ARTBA Bridge Report 2025 |
| Estimated cost of all identified bridge repairs | $467 billion | ARTBA Bridge Report 2025 |
Context note: ARTBA’s figures exclude Guam and the U.S. Virgin Islands and use an NBI download from June 24, 2025. The ARTBA 2025 report also ranks Iowa first with 4,424 poor bridges (18.7% of its inventory).
2. Age, Heavier Loads and Fatigue Exposure
Fatigue cracking is a function of time and load cycles, and US bridges have plenty of both. Approximately 45% of bridges have exceeded their planned design lives of 50 years (ASCE Infrastructure Report Card 2025), and ASCE warns that trucks heavier than the original design loads can overstress elements and cause metal fatigue and cracking. Replacement cannot catch up: at the 2014-2023 pace, a bridge built today would need to last 126 years.
Old steel is not only an American problem. Austria’s federal railway operator, whose decommissioned Pinkabach bridge is the field trial in section 6, runs over 8,700 bridges, about 1,450 of them steel, and many have reached 100 years of age (TUV Austria, EWGAE 2024). The full figures are in the ASCE 2025 bridges chapter.
| Metric | Value | Source |
|---|---|---|
| Average age of US bridges | About 47 years | ASCE Infrastructure Report Card 2025 |
| Share of bridges past their planned 50-year design life | Approximately 45% | ASCE Infrastructure Report Card 2025 |
| Bridges built between the 1800s and 1969 | 239,985 | ASCE Infrastructure Report Card 2025 |
| Bridges posted for load (weight-restricted) in 2024 | 63,085 of 623,218 | ASCE Infrastructure Report Card 2025 |
| Bridges replaced per year, 2014-2023, and implied service life | About 4,927 per year, implying 126 years | ASCE Infrastructure Report Card 2025 |
Trend note: the poor share of US bridges fell from 8.7% in 2014 to 6.8% in 2024, and poor deck area from 6.7% to 4.9% over the same decade (ASCE Infrastructure Report Card 2025). The share in good condition is falling too, which means more bridges are aging into fair condition, the band where fatigue details deserve the closest watch.
3. Fracture-Critical Members: The Inspection Rules and the Fern Hollow Lesson
A fracture critical member, called a nonredundant steel tension member (NSTM) in the current federal inspection rules, is a steel part carrying tension with no backup load path: if it breaks, part or all of the bridge can fall. That is why NSTMs get their own hands-on inspection clock, separate from the routine one, under the current federal text published on Cornell’s eCFR mirror of 23 CFR 650.311.
Fern Hollow shows what happens when the member is not even on the list. The NTSB found that PennDOT contractors did not identify the bridge legs, including the transverse tie plates, as fracture-critical members, so the legs did not consistently get hands-on inspections, and the City of Pittsburgh did not act on recommendations in inspection reports from 2005 to 2021 (NTSB, Fern Hollow Board Meeting Summary 2024). The NTSB board meeting summary adds that the asphalt wearing surface was nearly double the thickness recorded in inspection reports.
| Metric | Value | Source |
|---|---|---|
| Maximum hands-on inspection interval for an NSTM | 24 months | 23 CFR 650.311(c) |
| NSTM interval once its condition is coded 4 (poor) or lower | 12 months | 23 CFR 650.311(c) |
| Extended NSTM interval and its main conditions | Up to 48 months, only if built after 1978 under a fracture control plan, with no history of fatigue cracks and no pin and hanger assemblies | 23 CFR 650.311(c) |
| Longest routine interval under a risk-based (Method 2) policy | 72 months, with a service inspection midway | 23 CFR 650.311(a) |
| Average cost per bridge inspection assumed in the 2022 rule’s regulatory analysis | $2,000 (the proposed rule had assumed $257) | Federal Register, National Bridge Inspection Standards final rule 2022 |
| Fern Hollow Bridge length and fall height (January 28, 2022) | 447 feet long, fell about 100 feet | NTSB, Fern Hollow Board Meeting Summary 2024 |
| Action window had the legs been coded fracture-critical | Priority code 0, action within 7 days | NTSB, Fern Hollow Board Meeting Summary 2024 |
| New NTSB safety recommendations after Fern Hollow | 11 | NTSB, Fern Hollow Board Meeting Summary 2024 |
Inspection note: the NTSB also found that inspectors over a period of more than 15 years failed to clean corrosion before measuring, to accurately quantify remaining material and to recommend a structural review of the legs, which is why section 4 and section 5 look at how reliable visual inspection really is.
4. How Consistent Is Routine Visual Inspection?
Visual inspection is the main nondestructive evaluation (NDE, meaning testing without damaging the structure) method for bridges, yet no complete study of its reliability had been done since the inspection standards began in 1971 until a federal NDE Validation Center put 49 state inspectors through the same tasks. Only 3.9% of weld inspections correctly identified the presence of crack indications during the study’s in-depth tasks, and the study lists rushed inspections, fear of traffic, near visual acuity and light intensity among the factors tied to results (most recent available data: FHWA TechBrief FHWA-RD-01-105, Reliability of Visual Inspection for Highway Bridges, 2001).
The numbers are old, but they remain the national baseline for how much a condition rating depends on who holds the flashlight, and they explain why a 4 on one report can be a 6 on the next.
| Metric | Value | Source |
|---|---|---|
| Different condition ratings assigned to the same element, on average | Four or five | FHWA-RD-01-105, 2001 |
| Condition ratings predicted to fall within one point of the average | Only 68% | FHWA-RD-01-105, 2001 |
| Deck delamination surveys within 5 percentage points of the reference value | 6 of 22 teams | FHWA-RD-01-105, 2001 |
| Time spent on the same routine inspection from the ground | 12 to 82 minutes | Purdue JTRP 2019 |
5. Hands-On Visual Detection of Fatigue Cracks: The Purdue Benchmark
Probability of detection (POD) is the standard way to grade an inspection method: the share of real cracks of a given length that inspectors actually find. In POD language, a50 is the crack length found half the time and a90 the length found 90% of the time. Purdue’s S-BRITE center hung steel specimens with known fatigue cracks about 25 feet above the ground, told inspectors to treat them as fracture critical members, and sent 30 inspectors up in a man-lift, the closest thing to a real hands-on fracture-critical inspection. Results are in the JTRP executive summary.
For all 30 inspectors, the a90 crack length was 5-1/2 inches (Purdue JTRP 2019), and the data were too scattered to put a 95% confidence bound on it. Three cracks were missed by every inspector, the largest 3-1/4 inches long at a cover plate weld toe. The weathering-steel portion was stopped after 11 inspectors because rust formed across the static cracks, and 20 of the 21 hits there were made with a flashlight.
| Metric | Value | Source |
|---|---|---|
| Average detection rate from the ground (routine inspection) vs hands-on on weathering steel, 11 inspectors each | 4% vs 11% (21 hits in 198 observations; 6 of 11 inspectors found none on weathering steel) | Purdue JTRP 2019 |
| Average detection rate, hands-on, painted steel (30 inspectors, 70 cracks) | 65% (46 of 70), range 31% to 86% | Purdue JTRP 2019 |
| Crack length detected 50% of the time (a50) and 90% of the time (a90) | 1 inch and 5-1/2 inches (a90 of 4-1/2 inches for the 27 inspectors whose results fit the model) | Purdue JTRP 2019 |
| False calls per inspector on 147 painted specimens | Average 90, range 14 to 268 | Purdue JTRP 2019 |
| Change in probability of detection per extra inch of crack length | +16% (marginal effect 0.156) | Purdue JTRP 2019 |
Outlier note: the lowest hit-per-call rate was 13%, which the authors translate as 6.1 false calls for every crack correctly identified, and more false calls did not mean more hits. Detection also showed a slight negative trend with years of experience. Most recent available data: Purdue JTRP, 2019.
6. What Acoustic Emission Detects: Lab and Field Performance
Acoustic emission is passive: nothing is sent into the steel. Sensors on the surface wait for the brief stress wave a crack releases when it extends, then compare arrival times at several sensors to locate the source. The sensors work in the 50 kHz to 1 MHz range, and AE is not an effective method for detecting arrested cracks or other damage that is not progressing under load, so it answers a different question than a visual inspection: not “is there a crack” but “is this crack growing right now” (FHWA InfoTechnology, Acoustic Emission). For the arrival-time math behind source location, our acoustic camera localization statistics cover the same principle with microphones.
The best recent field-scale evidence comes from the Pinkabach railway bridge in Austria, which was shaken at about 4.2 Hz until notches grew into cracks. AE produced no indication in the first period (16,890 cycles), then flagged crack initiation during the second, and AE activity rose from 0.4 to 9.7 located events per cm per 1,000 fatigue cycles by the end of the test (TUV Austria, EWGAE 2024 paper). Across the nine monitored periods, which total 295,630 estimated cycles by our sum of the paper’s Table 1, 7 periods produced relevant indications. The two quiet periods were the very first and the one with the lowest stroke, when the stress range had been cut by more than a third.
| Metric | Value | Source |
|---|---|---|
| Pinkabach sensor layout and filter band | 3 location sensors plus 2 guard sensors (to reject outside noise), 95 to 300 kHz, 34 dB preamplifier | TUV Austria, EWGAE 2024 |
| Monitoring periods with relevant AE indications | 7 of 9 | TUV Austria, EWGAE 2024 |
| Peak AE amplitude at crack initiation vs late crack growth | About 72 dBAE vs up to 100 dBAE (channel saturation) | TUV Austria, EWGAE 2024 |
| AE signal classification accuracy on the test set, with K-SVD denoising | 92.56% damaged area, 94.27% undamaged area | PLoS ONE, Li et al. 2025 |
| Same classification without denoising, test set | 84.73% damaged area, 82.14% undamaged area | PLoS ONE, Li et al. 2025 |
| Locatable AE signal sets in the steel bridge deck specimen study | 1,473 (389 damaged region, 1,084 undamaged region), 8 sensors | PLoS ONE, Li et al. 2025 |
| Signal-to-noise ratio after denoising: K-SVD vs wavelet packet | 20.13 dB vs 15.24 dB | PLoS ONE, Li et al. 2025 |
| Minnesota field trial on cover-plated girders | Propagating crack detected on 2 of 3 bridges, extinguished crack on the third | MnDOT, MN/RC-1999-36, 1999 |
Context note: the PLoS ONE study was run on a 24 mm Q690q steel specimen in a 2,000 kN test machine, not a bridge in traffic, so its accuracy figures are an upper bound for field use. The 1999 Minnesota results come from the MnDOT report record and are the most recent available field data from that program.
7. Structural Health Monitoring in the Field: Deployments and Limits
Structural health monitoring (SHM) means leaving sensors on a structure to watch it continuously instead of waiting for the next inspection. The field record for fatigue cracks is small but instructive. On the Cedar Avenue Bridge, a fracture-critical tied arch in Minnesota, a 16-sensor AE system collected 22 months of field data, and a follow-up project installed two 16-sensor systems on a tie girder (UMN Center for Transportation Studies, Phase III).
Global vibration monitoring did not see the cracks at all: on the Pinkabach bridge, crack growth in the main girder flanges did not affect the structure’s global eigenfrequencies, while local methods (AE and distributed fiber optic sensing) tracked initiation and growth (EWSHM 2024 Pinkabach overview). That is the core argument for local sensing at known fatigue details. Vibration-based monitoring has its own uses, covered in our ground-borne vibration statistics.
| Metric | Value | Source |
|---|---|---|
| Cedar Avenue Bridge AE system and field record | 16 sensors, 22 months of field data | UMN CTS, AE Monitoring of a Fracture-Critical Bridge 2014 |
| Phase III setup and noise discrimination result | Two 16-sensor systems; noise never met all criteria in a set, every applicable fracture test met all criteria | UMN CTS, Protocols and Criteria for AE Monitoring 2015 |
| Railway crossbeams monitored at once in OBB’s 2017 proof of concept | 5 | TUV Austria, EWGAE 2024 |
| Pinkabach bridge test setup | Built 1952, 21.5 m span, 66 tons of ballast, cyclic stress of 70 N/mm2 | EWSHM 2024, Lachinger et al. |
| Effect of flange fatigue cracks on global eigenfrequencies | None detected | EWSHM 2024, Lachinger et al. |
| Distributed fiber optic strain sensing used on the crack zone | 0.65 mm resolution over 2.5 m of fiber at 62.5 Hz | EWSHM 2024, Lachinger et al. |
| Collapse load in the final static test (after cracks and 1.2 m web cuts) | 880 kN, roughly the weight of a locomotive | EWSHM 2024, Lachinger et al. |
Limit note: the EWSHM authors stress that AE measures relative crack growth only and gives no absolute crack length, while fiber optic sensing is still experimental, needs an expensive interrogator and is hard to install. In the 1999 Minnesota work, a bolted retrofit cut cracked-flange stress by 42% in the lab but only 43% and 8% on two field bridges (MnDOT 1999), a reminder that repairs also need monitoring.
Summary: Acoustic Emission and Bridge Fatigue Crack Monitoring by the Numbers
| Metric | Value | Source |
|---|---|---|
| US bridges rated poor, 2026 NBI table | 41,148 of 624,939 | FHWA, Bridge Condition by Highway System 2026 |
| Minnesota AE field trial on cover-plated girders | Propagating crack detected on 2 of 3 bridges | MnDOT, MN/RC-1999-36, 1999 |
| Daily crossings of poor bridges | 163 million | ARTBA Bridge Report 2025 |
| Cost of all identified bridge repairs | $467 billion | ARTBA Bridge Report 2025 |
| Bridges past their 50-year design life | Approximately 45% | ASCE Infrastructure Report Card 2025 |
| Implied service life at the current replacement pace | 126 years | ASCE Infrastructure Report Card 2025 |
| Maximum NSTM hands-on inspection interval | 24 months (12 if rated poor) | 23 CFR 650.311 |
| Fern Hollow years of unaddressed inspection recommendations | 2005 to 2021 | NTSB, Fern Hollow Board Meeting Summary 2024 |
| AE signal-to-noise ratio after denoising: K-SVD vs wavelet packet | 20.13 dB vs 15.24 dB | PLoS ONE, Li et al. 2025 |
| AE noise discrimination on a fracture-critical tied arch | Ambient noise never met all fracture criteria in a set | UMN CTS 2015 |
| Crack length found 90% of the time (a90) in hands-on inspection | 5-1/2 inches | Purdue JTRP 2019 |
| AE activity growth over the Pinkabach fatigue test | 0.4 to 9.7 located events per cm per 1,000 cycles | TUV Austria, EWGAE 2024 |
| Effect of flange fatigue cracks on global eigenfrequencies | None detected | EWSHM 2024, Lachinger et al. |
| AE classification test accuracy with denoising | 92.56% damaged, 94.27% undamaged | PLoS ONE, Li et al. 2025 |
| Cedar Avenue fracture-critical bridge AE deployment | 16 sensors, 22 months | UMN CTS 2014 |
Methodology and Sources
- FHWA, Bridge Condition by Highway System 2026 and 2025, National Bridge Inventory tables. Percentages are our calculations from the published counts and areas.
- ARTBA Bridge Report 2025, based on an NBI download of June 24, 2025.
- ASCE 2025 Infrastructure Report Card, Bridges chapter, citing FHWA data.
- 23 CFR 650.311, Inspection interval, current federal text.
- Federal Register, National Bridge Inspection Standards final rule, May 6, 2022.
- NTSB, Collapse of the Fern Hollow Bridge, Board Meeting Summary, February 21, 2024 (HWY22MH003; final report HIR-24-02).
- FHWA NDE Validation Center, Reliability of Visual Inspection for Highway Bridges, TechBrief FHWA-RD-01-105, 2001.
- Purdue JTRP, Probability of Detection Study for Visual Inspection of Steel Bridges, Volume 1, FHWA/IN/JTRP-2019/21.
- FHWA InfoTechnology, Bridge: Acoustic Emission (AE).
- Lackner, Marihart and Prokofyev (TUV Austria), Monitoring of a fatigue test on a full-size railway steel bridge with acoustic emission, EWGAE 2024.
- Lachinger et al., Destructive Testing of the Pinkabach Railway Bridge, EWSHM 2024.
- Li et al., Fatigue micro-crack identification of steel bridge decks based on acoustic emission, PLoS ONE 2025.
- McKeefry and Shields, Acoustic Emission Monitoring of Fatigue Cracks in Steel Bridge Girders, MnDOT MN/RC-1999-36.
- University of Minnesota Center for Transportation Studies, Advanced Warning System for Fracture Critical Steel Bridges, Phase II and Monitoring of Fracture Critical Steel Bridges, Phase III, with the 2014 report Acoustic Emission Monitoring of a Fracture-Critical Bridge (MnDOT 2014-15) and the 2015 report Protocols and Criteria for Acoustic Emission Monitoring of Fracture-Critical Steel Bridges.
- Data watch: the national count of fracture-critical (NSTM) bridges is not part of the condition tables used here, and the figures quoted in press coverage could not be traced to a primary federal table in this session, so no NSTM count is used. ARTBA’s 2025 poor count (41,677) and the June 2025 NBI table (41,685) differ slightly because of territory coverage and download date. The 2001 visual inspection study, the Minnesota AE field trial (1999) and the Purdue POD study (2019) are older than three years and are the most recent available data of their kind. AE accuracy figures from laboratory specimens (PLoS ONE 2025) should not be read as field detection rates, and the Pinkabach test used artificial notches on a decommissioned bridge. ARTBA’s 2026 bridge report, based on the 2026 NBI data, is expected soon and should update the crossings and repair-cost figures.
Last updated: October 25, 2026. We update this roundup quarterly, and the next refresh is expected when ARTBA publishes its 2026 Bridge Report on the 2026 National Bridge Inventory data.