Unplanned equipment downtime now costs the world’s 500 largest companies $1.4 trillion a year, equal to 11% of their revenue (Siemens Senseye, The True Cost of Downtime 2024). That bill is the economic case for listening to machines: a failing bearing, a leaking valve or a loose belt changes its sound long before it stops a line. Machine sound is finally being benchmarked in the open, too: the top DCASE 2026 anomalous sound detection system scored 70.24 against a 59.80 baseline, while 57% of manufacturers now report predictive maintenance as a deployed AI use case (Augury State of Production Health 2026). We aggregated data from Siemens, ABB, the DCASE Challenge (IEEE AASP), Augury, Forrester, Deloitte, the U.S. Department of Energy, IMARC Group, MarketsandMarkets and Mordor Intelligence. For the wider connected-device picture behind these sensors, see our IoT statistics roundup.
TL;DR
- $1.4 trillion a year lost to unplanned downtime by the Fortune Global 500, 11% of revenue (Siemens Senseye, True Cost of Downtime 2024)
- $2.3 million per hour of downtime in automotive plants, more than $600 a second (Siemens Senseye, True Cost of Downtime 2024)
- About $125,000 per hour is the typical cost of an unplanned outage across 3,215 surveyed businesses (ABB, Value of Reliability survey 2023)
- $15.60 billion predictive maintenance market in 2025, heading to $40.46 billion in 2030 (IMARC Group, Predictive Maintenance Market 2025)
- 11.62% CAGR for ultrasound emission monitoring to 2031, one of the fastest-growing techniques (Mordor Intelligence, Machine Condition Monitoring Market 2026)
- 70.24 official score for the top DCASE 2026 Task 2 system vs 59.80 for the MSE baseline (DCASE 2026 Challenge results)
- 119 submissions from 35 teams in DCASE 2025 Task 2, the highest count of the 2020 to 2025 editions (DCASE 2025 Task 2 description paper)
- 57% of manufacturers have deployed predictive maintenance, the most-deployed AI use case (Augury State of Production Health 2026)
- 42% have scaled AI across more than half their sites, up from 14% a year earlier (Augury State of Production Health 2026)
- 310% three-year ROI and payback under 6 months for AI machine health monitoring (Forrester TEI of Augury 2025)
- 20% to 30% of compressor output can be wasted by compressed air leaks, the classic ultrasonic detection use case (U.S. DOE Compressed Air Sourcebook)
- 47% of manufacturers cite poor data quality as the top barrier to AI maturity (Augury State of Production Health 2026)
1. The Cost of Not Listening: Unplanned Downtime
Downtime keeps getting more expensive per hour, which changes the math for any monitoring system. Siemens’ 2024 analysis puts the annual loss for the Fortune Global 500 at $1.4 trillion, up from $864 billion (8% of revenue) in 2019 to 2020, a 62% increase on our calculation ($1.4 trillion divided by $864 billion = 1.62). When a single automotive line burns more than $600 a second, even a few minutes of earlier warning from an acoustic or vibration sensor has a measurable cash value.
The ABB data tells the same story from the shop floor: outages are routine, not exceptional. Two-thirds of industrial businesses still see at least one unplanned stoppage a month, and one in five still runs equipment to failure. At roughly $125,000 an hour, the business case for earlier warning does not need a Fortune 500 balance sheet. You can read the methodology in the Siemens True Cost of Downtime 2024 report and in ABB’s Value of Reliability release.
| Metric | Value | Source |
|---|---|---|
| Annual cost of unplanned downtime, Fortune Global 500 | $1.4 trillion (11% of revenue) | Siemens Senseye, True Cost of Downtime 2024 |
| Same cost in 2019 to 2020 | $864 billion (8% of revenue) | Siemens Senseye, True Cost of Downtime 2024 |
| Cost per hour of downtime, automotive | $2.3 million (over $600 per second) | Siemens Senseye, True Cost of Downtime 2024 |
| Cost per hour of downtime, fast-moving consumer goods | $36,000 | Siemens Senseye, True Cost of Downtime 2024 |
| Hours lost per month at an average large plant | 27 hours | Siemens Senseye, True Cost of Downtime 2024 |
| Heavy industry hourly downtime cost vs 2019 | 4x higher | Siemens Senseye, True Cost of Downtime 2024 |
| Typical hourly cost of unplanned outage (global / U.S.) | ~$125,000 / ~$103,000 | ABB, Value of Reliability survey 2023 |
| Industrial businesses with unplanned outages at least monthly | 68% | ABB, Value of Reliability survey 2023 |
Context note: the Siemens figures model the Fortune Global 500, so they skew toward the largest asset bases. ABB’s survey was fielded by Sapio Research in July 2023 among 3,215 plant maintenance decision-makers across 11 sectors, which makes it a better proxy for a typical mid-size plant.
2. Predictive Maintenance Market Size and Growth
Every analyst firm agrees the predictive maintenance market is growing at double digits, and almost none agree on its size. Base-year estimates (2024 to 2026) range from about $9.7 billion to $15.6 billion, and growth rates span 11.4% to 28.7% a year, because some firms count only software and analytics while others fold in sensors, services and asset performance platforms. The spread is a reminder that acoustic sensing is one input into a much larger maintenance stack, not a market of its own yet.
The most conservative recent forecast comes from MarketsandMarkets, which sees $13.89 billion in 2026 growing to $23.79 billion by 2031, with sensors and sensing devices as the largest infrastructure component. IMARC Group is far more bullish, with $91.04 billion by 2034.
| Metric | Value | Source |
|---|---|---|
| Predictive maintenance market, 2025 | $15.60 billion | IMARC Group, Predictive Maintenance Market 2025 |
| Predictive maintenance market, 2030 / 2034 | $40.46 billion / $91.04 billion (21.01% CAGR) | IMARC Group, Predictive Maintenance Market 2025 |
| North America share of predictive maintenance, 2025 | 38.9% | IMARC Group, Predictive Maintenance Market 2025 |
| On-premises share of deployments, 2025 | 57.3% | IMARC Group, Predictive Maintenance Market 2025 |
| Predictive maintenance market, 2026 to 2031 | $13.89 billion to $23.79 billion (11.4% CAGR) | MarketsandMarkets, Predictive Maintenance Market 2026 |
| Predictive maintenance market, 2024 to 2030 | $12.73 billion to $56.21 billion (28.69% CAGR) | Research and Markets, Predictive Maintenance Market Outlook 2030 |
| Predictive maintenance market, 2024 to 2030 | $10.93 billion to $41.90 billion | MarkNtel Advisors, Global Predictive Maintenance Market |
Outlier note: a MarketsandMarkets release distributed via PR Newswire on the same date gives a smaller base ($9.71 billion in 2026 to $16.74 billion by 2031, 11.5% CAGR). Treat the range, not any single point, as the market size.
3. Acoustic and Ultrasound Monitoring Segments
Sound-based monitoring is a niche inside the niche, but it is the faster-growing niche. Ultrasound emission monitoring is forecast to grow at 11.62% a year to 2031, outpacing the 9.47% growth of machine condition monitoring overall (Mordor Intelligence). Ultrasound catches what vibration often misses at an early stage: small leaks, electrical discharge and the first friction of a dry bearing.
Vibration analysis still holds the largest share of condition monitoring at 33.75%, so microphones and ultrasonic sensors are mostly additive rather than a replacement. The other signal in this data is wireless retrofits: battery IIoT sensors are the fastest-growing deployment model, which is exactly the form factor most acoustic anomaly detection products ship in. Edge inference matters here too, which is covered in our edge AI statistics. Figures below come from Mordor Intelligence’s machine condition monitoring report and its acoustic emission testing report.
| Metric | Value | Source |
|---|---|---|
| Machine condition monitoring market, 2026 to 2031 | $1.53 billion to $2.41 billion (9.47% CAGR) | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Ultrasound emission segment growth to 2031 | 11.62% CAGR | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Vibration analysis share of condition monitoring, 2025 | 33.75% | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Wireless IIoT sensor networks growth | 12.22% CAGR | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Online / continuous systems share, 2025 | 47.55% | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Oil and gas share of condition monitoring demand, 2025 | 27.05% | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Acoustic emission testing market, 2026 to 2031 | $0.42 billion to $0.55 billion (5.34% CAGR) | Mordor Intelligence, Acoustic Emission Testing Market 2026 |
| Acoustic emission testing: sensors share / software growth | 41.92% share / 8.45% CAGR | Mordor Intelligence, Acoustic Emission Testing Market 2026 |
Context note: Mordor’s condition monitoring definition is narrow (about $1.4 billion in 2025), so it is not directly comparable with the broader predictive maintenance totals in Section 2. In acoustic emission testing, oil and gas leads with 24.35% of demand and Asia-Pacific holds 34.05% of revenue.
4. DCASE Benchmarks: How Good Is Machine Sound Anomaly Detection?
DCASE Task 2 is the closest thing the field has to a public scoreboard, and it is deliberately hard. Systems are trained only on normal sounds, then judged on anomalies they have never heard, increasingly on machine types they have never heard either. The DCASE 2026 winner scored 70.24 on the official metric, while the simple MSE autoencoder baseline scored 59.80 and still ranked 68th, which says more about how tight the field is than about weak baselines.
Participation has grown with the difficulty. After 117 submissions in the first edition (2020), the task dipped to 75 in 2021 when domain shift was introduced, then climbed back to a high of 119 submissions from 35 teams in 2025. The official score is a harmonic mean of AUC and partial AUC (pAUC at a false-positive rate of 0.1) across machine types, sections and domains. See the DCASE 2025 Task 2 paper and the DCASE 2026 results page.
| Metric | Value | Source |
|---|---|---|
| DCASE 2020 Task 2 participation | 117 submissions, 40 teams | DCASE 2020 Task 2 description paper (arXiv 2006.05822) |
| DCASE 2021 / 2022 participation | 75 submissions, 26 teams / 81 submissions, 31 teams | DCASE 2021 and 2022 Task 2 papers (arXiv 2106.04492, 2206.05876) |
| DCASE 2023 / 2024 participation | 86 submissions, 23 teams / 96 submissions, 27 teams | DCASE 2023 and 2024 Task 2 papers (arXiv 2305.07828, 2406.07250) |
| DCASE 2025 participation | 119 submissions, 35 teams | DCASE 2025 Task 2 description paper (arXiv 2506.10097) |
| DCASE 2024 top system vs MSE baseline | 66.24 vs 56.51 | DCASE 2024 Task 2 results |
| DCASE 2025 top system vs baselines (Mahalanobis / MSE) | 61.63 vs 56.51 / 54.43 | DCASE 2025 Task 2 results |
| DCASE 2026 top system vs baselines (MSE / Mahalanobis) | 70.24 vs 59.80 / 54.76 | DCASE 2026 Task 2 results |
| Rank of DCASE 2026 baselines in the system table | 68th (MSE) and 137th (Mahalanobis) | DCASE 2026 Task 2 results |
Context note: scores are not comparable across years because each edition evaluates different machine types and conditions. In 2025, the top three teams finished within one point of each other (61.63, 61.57 and 61.20), per the DCASE 2025 results.
5. Adoption: Predictive Maintenance Leaves the Pilot Stage
The adoption question has flipped from “whether” to “how many sites.” 42% of manufacturers have now scaled AI across more than half their facilities, up from 14% a year earlier, and predictive maintenance is the single most-deployed AI use case at 57% (Augury State of Production Health 2026, 501 leaders in the U.S., U.K., Germany and France). Predictive maintenance adoption alone rose 22 percentage points year over year.
Deloitte’s larger U.S. sample is more cautious: fewer than a third of large manufacturers run AI or machine learning at facility or network scale. The gap between the two surveys is mostly a sample effect (Augury surveys a population already interested in production health), but both point the same way. The details are in Augury’s 2026 release and Deloitte’s 2025 Smart Manufacturing survey.
| Metric | Value | Source |
|---|---|---|
| Manufacturers with predictive maintenance deployed | 57% | Augury State of Production Health 2026 |
| Year-over-year rise in predictive maintenance adoption | +22 percentage points | Augury State of Production Health 2026 |
| Manufacturers scaling AI across more than half of sites | 42% (vs 14% a year earlier) | Augury State of Production Health 2026 |
| Manufacturers planning to increase AI investment in 2026 | 83% | Augury State of Production Health 2026 |
| Large manufacturers using AI/ML at facility or network level | 29% | Deloitte 2025 Smart Manufacturing and Operations Survey |
| Large manufacturers piloting AI/ML | 23% | Deloitte 2025 Smart Manufacturing and Operations Survey |
| Manufacturers prioritizing IIoT investment in the next 24 months | 27% | Deloitte 2025 Smart Manufacturing and Operations Survey |
| Industrial businesses still relying on run-to-fail maintenance | 21% | ABB, Value of Reliability survey 2023 |
Context note: Deloitte surveyed 600 executives at U.S. manufacturers with $500 million or more in revenue and 1,000 or more employees, fielded August to September 2024. Augury’s survey was fielded in March 2026 by Endeavor Business Intelligence with IndustryWeek.
6. ROI and Use Cases: Where Listening Pays Back
The best-documented returns come from rotating equipment and compressed air, two areas where sound carries early fault information. A Forrester Total Economic Impact study found a 310% three-year ROI and payback in under six months for AI machine health monitoring at a $20 billion composite manufacturer, driven mostly by $16.8 million in avoided downtime. The study was commissioned by the vendor (Augury) and built on four customer interviews, so treat it as an upper-range benchmark rather than an average.
Compressed air is the textbook acoustic use case because leaks emit ultrasound that is inaudible on a noisy floor. The U.S. Department of Energy’s compressed air sourcebook notes that leaks can waste 20% to 30% of a compressor’s output, and that a proactive detection and repair program can cut that to under 10%. The full Forrester breakdown is on the TEI study page.
| Metric | Value | Source |
|---|---|---|
| Three-year ROI / NPV of AI machine health monitoring | 310% / $20.1 million | Forrester TEI of Augury Machine and Process Health 2025 |
| Payback period | Under 6 months | Forrester TEI of Augury Machine and Process Health 2025 |
| Unplanned downtime avoided per site | 95 hours ($16.8 million over three years) | Forrester TEI of Augury Machine and Process Health 2025 |
| Maintenance cost reduction per asset / throughput gain per line | 15% / 5% | Forrester TEI of Augury Machine and Process Health 2025 |
| Total three-year benefits vs costs (present value) | $26.6 million vs $6.5 million | Forrester TEI of Augury Machine and Process Health 2025 |
| Compressor output wasted by leaks | 20% to 30% (target: under 10%) | U.S. DOE, Improving Compressed Air System Performance Sourcebook |
| Businesses reporting maintenance increased uptime in the past year | 92% (38% gained 25% or more) | ABB, Value of Reliability survey 2023 |
| Production output gains reported after smart manufacturing rollouts | 10% to 20% | Deloitte 2025 Smart Manufacturing and Operations Survey |
Context note: the oft-quoted figures that predictive maintenance delivers a 10x ROI, 25% to 30% lower maintenance cost and 70% to 75% fewer breakdowns come from the U.S. DOE Operations and Maintenance Best Practices Guide, Release 3.0. Most recent available data: U.S. DOE, 2010. The underlying studies are older still and hard to trace, so we list them for context only.
7. Challenges: Data Quality, Noise and Domain Shift
The barriers have moved from budget to data. Poor data quality is now the top roadblock to AI maturity, cited by 47% of manufacturers and up 20 points in a year (Augury State of Production Health 2026). For acoustic systems the data problem is physical as well as organizational: factory noise, microphone placement and changing operating conditions all shift the sound of a healthy machine.
The DCASE results quantify that difficulty. In 2024, only 10 of 27 teams beat the simple autoencoder baseline on unseen machines, and in 2025 the organizers reported that development-set performance did not predict evaluation results. The 2026 task responded by adding a second, distant microphone so systems can separate environmental noise from the machine itself, per the DCASE 2026 task description. Market consolidation is a further risk: AWS closed Amazon Monitron to new customers in 2024. Security is the other open front, covered in our industrial IoT security statistics.
| Metric | Value | Source |
|---|---|---|
| Manufacturers citing poor data quality as top AI barrier | 47% (up 20 points year over year) | Augury State of Production Health 2026 |
| Manufacturers citing lack of proven value or defined use cases | 39% | Augury State of Production Health 2026 |
| Manufacturers citing workforce constraints / unplanned downtime as top challenge | 43% / 40% | Augury State of Production Health 2026 |
| Large manufacturers struggling to fill production and operations roles | 48% | Deloitte 2025 Smart Manufacturing and Operations Survey |
| DCASE 2024 teams beating the autoencoder / Mahalanobis baseline | 10 of 27 / 11 of 27 | DCASE 2024 Task 2 description paper (arXiv 2406.07250) |
| DCASE 2025 teams beating both baselines | 20 of 35 | DCASE 2025 Task 2 description paper (arXiv 2506.10097) |
| Normal training clips per section: source vs target domain | 990 vs 10 | DCASE 2025 and 2026 Task 2 dataset descriptions |
| Amazon Monitron closed to new customers / device sales ended | October 31, 2024 / July 2025 | AWS Machine Learning Blog, October 2024 |
Context note: the 990-versus-10 clip split is the core of the domain-shift problem. A deployed system usually has months of audio from its original setting and only a handful of clips after a machine is moved, retuned or surrounded by new equipment. Noise itself is a broader industrial health issue as well, see our noise pollution statistics.
Summary: Industrial Acoustic Anomaly Detection by the Numbers
| Metric | Value | Source |
|---|---|---|
| Annual unplanned downtime cost, Fortune Global 500 | $1.4 trillion (11% of revenue) | Siemens Senseye, True Cost of Downtime 2024 |
| Automotive downtime cost per hour | $2.3 million | Siemens Senseye, True Cost of Downtime 2024 |
| Hours lost per month at an average large plant | 27 hours | Siemens Senseye, True Cost of Downtime 2024 |
| Typical hourly cost of unplanned outage | ~$125,000 | ABB, Value of Reliability survey 2023 |
| Businesses with monthly unplanned outages | 68% | ABB, Value of Reliability survey 2023 |
| Predictive maintenance market, 2025 | $15.60 billion | IMARC Group, Predictive Maintenance Market 2025 |
| Predictive maintenance market, 2031 | $23.79 billion | MarketsandMarkets, Predictive Maintenance Market 2026 |
| Machine condition monitoring market, 2026 | $1.53 billion | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Ultrasound emission monitoring growth | 11.62% CAGR to 2031 | Mordor Intelligence, Machine Condition Monitoring Market 2026 |
| Acoustic emission testing market, 2031 | $0.55 billion | Mordor Intelligence, Acoustic Emission Testing Market 2026 |
| DCASE 2026 top score vs MSE baseline | 70.24 vs 59.80 | DCASE 2026 Task 2 results |
| DCASE 2025 submissions / teams | 119 / 35 | DCASE 2025 Task 2 description paper |
| DCASE 2020 submissions / teams | 117 / 40 | DCASE 2020 Task 2 description paper |
| Manufacturers with predictive maintenance deployed | 57% | Augury State of Production Health 2026 |
| Manufacturers scaling AI across more than half of sites | 42% | Augury State of Production Health 2026 |
| Large manufacturers using AI/ML at scale | 29% | Deloitte 2025 Smart Manufacturing and Operations Survey |
| Three-year ROI of AI machine health monitoring | 310% | Forrester TEI of Augury 2025 |
| Compressor output wasted by air leaks | 20% to 30% | U.S. DOE Compressed Air Sourcebook |
| Poor data quality as top AI barrier | 47% | Augury State of Production Health 2026 |
| Businesses still on run-to-fail maintenance | 21% | ABB, Value of Reliability survey 2023 |
Methodology and Sources
- Siemens Senseye, The True Cost of Downtime 2024: report PDF. The 27-hours-per-month and 4x heavy-industry figures were confirmed through Etteplan’s summary of the report.
- ABB, Value of Reliability survey (Sapio Research, July 2023, 3,215 respondents): ABB press release and Canadian Manufacturing coverage for the uptime and run-to-fail breakdown.
- IMARC Group, Predictive Maintenance Market: report page.
- MarketsandMarkets, Predictive Maintenance Market (March 2026): press release and the PR Newswire version.
- Research and Markets, Predictive Maintenance Market Outlook 2030: report page.
- MarkNtel Advisors, Global Predictive Maintenance Market: press release.
- Mordor Intelligence: Machine Condition Monitoring Market and Acoustic Emission Testing Market.
- DCASE Challenge Task 2 (IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events): 2024 results, 2025 results, 2026 task description and 2026 results.
- DCASE Task 2 description papers on arXiv: 2020, 2021, 2022, 2023, 2024, 2025 and 2026.
- Augury, State of Production Health 2026 (501 respondents, March 2026): press release.
- Forrester Consulting, The Total Economic Impact of Augury Machine and Process Health (2025, commissioned by Augury): study page.
- Deloitte, 2025 Smart Manufacturing and Operations Survey (600 executives): press release.
- U.S. Department of Energy: Improving Compressed Air System Performance Sourcebook and Operations and Maintenance Best Practices Guide, Release 3.0.
- AWS Machine Learning Blog, Maintain access and consider alternatives for Amazon Monitron (October 1, 2024): post.
- Data watch: predictive maintenance market sizes conflict sharply (2025 to 2026 baselines from about $9.7 billion to $15.6 billion, CAGRs from 11.4% to 28.7%) because firms define scope differently, and MarketsandMarkets published two different baselines on the same date. No major analyst firm publishes a stand-alone market size for acoustic anomaly detection, so acoustic share must be inferred from the ultrasound and acoustic emission segments. DCASE scores are not comparable across editions. The ABB survey dates from 2023 and the DOE compressed air guidance from 2016 (most recent available data), and the DOE O&M Guide figures from 2010 are cited for context only. The Forrester ROI study was vendor-commissioned. We looked for ARC Advisory Group asset performance management data but found no public figures we could verify.
Last updated: October 3, 2026. We update this roundup quarterly, and the next refresh is expected when the DCASE 2026 Workshop proceedings and the Task 2 results analysis paper are published.