The Korean study said to prove the AML Pro works, taken apart section by section

AWhy the device cannot do what it is sold on

First, the device itself. The AML Pro, made by SubSurface Instruments, is sold on finding plastic pipe “up to 20 feet” down, and on working “in clay, wet soil, snow or even standing water”. Five findings, each set out in full in our earlier work:

1. It works at the microwave oven frequency. 2.45 GHz is used in microwave ovens because it gets a few centimetres into wet food before it is absorbed. The signal loses about two thirds of its strength every 1.4 cm in water and every 3 cm in moist clay. By 30 cm of moist clay it is twenty-two thousand times weaker, one way. The working.
2. The signal cannot make the round trip. A round trip to a pipe at 6 m in moist clay costs roughly 3,476 dB in absorption alone (every 10 dB is a further factor of ten). Even at the highest power the rules allow for this band, a receiver can recover about 154 dB. The gap is about 3,300 dB. No antenna or signal processing can close a gap that size. The link budget.
3. Its power is low, and it is not stated in the paperwork. Under the Radio Equipment Regulations 2017, radio equipment sold in Great Britain has to state its maximum transmitted power in the instructions supplied with it. The AML Pro’s English instructions do not, and neither did the manual in the box of the unit we bought. The only power figure we have found in any edition of the manual, 10 mW, is on the German page of the 2014 manual and on no other language page. Worked out from the manufacturer’s own certification laboratory measurement, it is about 41 mW, roughly what a wireless keyboard is allowed to transmit, and about 24 dB below the figure the link budget above assumed, which makes the gap larger still. The power piece.
4. Radar responds to differences in electrical properties. The “density” language is the brochure’s. A buried plastic pipe in soil offers little electrical difference and a weak reflection at the best of times. What radar detects.
5. The indicator follows the operator’s hand. The manufacturer’s own FAQ says the lights come on and go off “approximately every 2 inches”. Half the wavelength of 2.45 GHz is 2.4 inches. So the lights change with small movements of the hand, and the manuals tell the operator how to make those movements. The earliest edition told an operator who could “feel the target should be there” to “try twisting your wrist”. The manual pages.

In fourteen years on sale we have found no published blinded test, by anyone, showing it finds a hidden pipe better than chance. The inventor of the underlying technology told us he tested it on pipe at one to two feet. His account. The patent it was sold on lapsed in 2023, and, despite the sales story, no NASA lunar instrument ever flew at its frequency. The record.

BThe paper, section by section

Lee, S.-J. and Yun, H.-S. (2025), “Performance Evaluation of AML Equipment for Determining the Depth and Location of Subsurface Facilities in South Korea”, Applied Sciences 15(11), 5794, MDPI, doi 10.3390/app15115794. Geodesy Laboratory, Sungkyunkwan University. Special issue on ground penetrating radar. Received 18 April 2025, published 21 May 2025. It is free to read, and I would encourage anyone weighing this device to read it alongside what follows.

Abstract

The 1.5 m figure for “most GPR systems” is never supported in the paper. No systems are named, no numbers are given, no data are shown. The closing sentence is the one a buyer is most likely to see, and it is a claim about “regulated environments” in general, drawn from one test site in ground the paper itself calls ideal.

This is the sentence that matters most to a buyer. The test site is not an ordinary research field. Korean law requires every pipe and cable locator held by a registered utility surveying firm to pass a government performance inspection every three years. In May 2022, Korea’s transport ministry said the Sungkyunkwan University surveying technology centre was the only body in the country registered to carry out that inspection for locators. From that university, “meets South Korea’s legal requirements” will read to many buyers as a pass. The paper does not say the AML was ever put through the formal inspection, and its own Table 2 (Section 3.2 below) prints three depth readings outside the limit it applies.

1. Introduction

The paper’s first example of why utility detection matters is “the 2003 Daegu subway fire disaster in South Korea, which led to over 100 fatalities”. That fire was an arson attack on a train, not a failure to find a buried service. The Daegu disaster caused by striking a buried gas pipe was a different one, in 1995. It is a small error, but it is the paper’s opening example.

It then describes the device:

The AML does not use acoustic or ultrasonic detection; the paper itself says so later. And the “preliminary studies” are reference 10, the manufacturer’s own product page (which now returns “page not found”), and reference 11, a 2024 paper on finding defects inside pipes with ultrasonic guided waves. Neither is a study of the AML. There is no evidence for “greater depth penetration” behind that sentence.

2.1 Ground-Penetrating Radar

This is correct, and it works against the paper’s own claim. The AML works at 2,450 MHz, above that whole range. By the paper’s own rule, it would reach less deep than the GPR it is compared with.

2.2 GPR performance under identical site conditions

In this ground, a GPR tried at several antenna frequencies and with signal processing could not see the polyethylene (PE) pipe, and the authors say why: the ground absorbs the signal, and the plastic reflects little of it. Absorption is greater at higher frequencies, the pipe reflects no better, and the AML transmits about 41 mW. The section then says the AML “successfully detected the same 3.0 m deep PE pipeline with high accuracy”, and offers no mechanism for how. The GPR is not named, and the paper does not say which frequencies it used; the range it gives for GPR is 10 MHz to 1000 MHz, all below the AML’s 2,450 MHz.

2.3 AML System

The paper describes how the AML works in four different ways:

• “ultra-high-frequency electromagnetic sensing and acoustic/ultrasonic detection” (Section 1)
• “ultrasonic-based detection systems like the AML”
• “the device emits high-frequency acoustic- or radio-frequency waves into the ground”, with depth taken from the time delay of the echo
• “While the AML system operates using ultra-high radio frequency (UHRF) sensing and not acoustic wave propagation”

Ultrasound is a form of sound, so the device cannot be ultrasonic and “not acoustic” at the same time. Nor does the AML time an echo. The paper’s own Figure 4, credited to the manufacturer, shows the real method: estimating depth “at a 45° angle”. The manufacturer’s manual spells it out: tilt the handle down to 45 degrees and walk backwards until the pipe is detected again. Depth comes from how far the operator walks, not from timing an echo.

The power is missing here too. A paper evaluating a radio device, in a special issue on radar, never states how much power the device transmits. Together with the frequency and the ground itself, it sets how deep a radio signal can reach and return. It is not in the paper, it is not in the manufacturer’s English instructions, and it was not in the paperwork supplied with the unit we bought in the UK. Worked out from the manufacturer’s own certification file, it is about 41 mW. How we worked it out. Whether the authors had that figure, and whether it came with their unit, the paper does not say.

The section’s description of what the device can do matches its sellers’ wording: that it finds “plastic, metal, wood, cable, or pipe” and works in “clay, wet soil, snow, and even standing water”, “using GPS and advanced patented digital signal processing”. The paper says this is documented in “the manufacturer’s specifications and product manuals, as well as in third-party technical reports”, citing:

None is a third-party technical report. “Patented” was also out of date when the paper was written: the matching patent, US 7,898,456, lapsed on 3 April 2023 for unpaid fees.

2.4 Establishment of the performance test site

Five features of the test limit what it can show:

• One operator. For a handheld whose indicator follows the operator’s movement, a single operator cannot separate the device from the person. The evidence for this device.
• Known positions. Every pipe was surveyed in 3D with a total station (a surveying instrument) before testing. The paper does not say the operator was kept from knowing where they were.
• The same site, run since 2010 by the authors’ own university centre. According to Korea’s transport ministry in May 2022, that centre was the only body in Korea registered to carry out the legally required performance inspection of pipe locators. The layout is fixed and documented.
• Dry ground by design. No rain for three days before every session, in weathered (decomposed) granite soil the paper calls “favorable”.
• Aids on the testbed. The paper says the site includes “high-contrast positioning for non-metallic lines” and conductive tracer wires on the metallic ones, without saying what the positioning aids are. Markers of that kind would help an operator who can see them.

The paper describes no blind trial, no second operator and no run over empty ground. It later says “no false positives or spurious detections were observed”, but with no empty-ground control the test had no way to reveal one.

2.5 Testing of various instruments

This section repeats 2.4 almost word for word: the same site description, the same 25 m by 25 m testbed, the same backfill, the same total station. Apart from a flowchart (Figure 6), it adds nothing about how the AML readings were taken or recorded.

3.1 Summary of field test results

The paper describes the targets as pipes “installed horizontally at controlled depths of 2 m and 3 m” and focuses on “the 3 m deep non-metallic targets”. Table 1, headed “RCP (3 m)” and “PE Pipe (3 m)”, gives the reference depths:

None of the twenty is 3 m. The text then describes 2.630 m and 2.524 m as “the average depth of the detected signals”, although these are the surveyed reference depths, not AML readings. The table also has visible slips: the concrete position at point 7 reads 2.816 in a run that otherwise climbs from 2.180 to 2.191, and the last polyethylene depth, 2.6115, is out of step with the rest.

3.2 Quantitative assessment against the Korean regulations

Table 2, printed a few lines below, gives these polyethylene depth deviations:

The section also says both horizontal and vertical deviations must be within ±0.30 m. The paper’s own Table 3, the Korean standard, says ±20 cm horizontal and ±30 cm depth. Table 3 is also titled “GPR equipment performance evaluation criteria”: the standard the paper applies was written for GPR.

3.3 Statistical analysis

For each of its four measures the paper gives the average error, the spread (standard deviation) and the RMSE, from the same 320 readings. Those three are tied together: RMSE squared equals the average squared plus the spread squared. So the RMSE can never be smaller than that combination.

For position the three numbers agree: a gap of three ten-thousandths of a metre is rounding. For depth, which is what matters on a dig, the three figures cannot all come from the same readings.

Over 320 readings, random errors average close to zero. These average +0.0822 m and +0.0975 m. With a spread of about 0.12 m over 320 readings, chance alone would move the average by about 0.007 m. The published averages are 12 and 15 times that: a consistent lean in one direction, which is what systematic bias means. And if the errors really followed a bell curve with that average and spread, about 12 of every 320 readings would fall outside ±0.30 m, which is what Table 2 shows, although Section 3.2 says otherwise.

The section ends by calling the device capable of “sub-decimeter accuracy”. A decimetre is 10 cm. The paper’s own depth RMSE figures are 12.3 cm and 12.7 cm.

4. Discussion

This paragraph limits what the abstract can claim. The AML Pro is sold on clay, wet soil and standing water. The authors say their result may not hold there, and they tested in none of it. The discussion also admits the testbed has nothing deeper than 3 m, so no depth beyond that was tested either, against a device marketed at 20 feet.

5. Conclusions

The conclusions repeat “sub-decimeter accuracy” and call the AML “a practical and efficient alternative to traditional GPR- and EMI-based methods”, with “significant potential to improve excavation safety”. None of that follows from one operator, one site, dry ground, known pipes, an unstated power, and the only two sessions printed, which break the stated limit.

Back matter and corrections

• Funding: a National Research Foundation of Korea grant. Conflicts: none declared.
• Data: “The original contributions presented in this study are included in this article”. They are not: only two of the 32 sessions are printed.
• Correction: the paper has been republished “to resolve spelling errors”. None of the problems above was changed.

CWhat I have asked, and the right of reply

I have put these questions to the authors:

1. What maximum radio-frequency power did the unit transmit, did you know it at the time, and was it stated in the manual or paperwork supplied with the unit?
2. How does a 2.45 GHz signal at that power reach a pipe that your GPR could not see in the same ground, and at what frequencies was that GPR operated?
3. How were the depth RMSE figures calculated, given that they are lower than the reported averages and spreads allow?
4. Why does Section 3.2 say every deviation was within ±0.30 m when Table 2 prints three that were not?
5. Did the operator know where the pipes were, and was any blind or empty-ground trial run?
6. Where did the AML unit come from, and was the manufacturer or a distributor involved in supplying it, training the operator or the testing?
7. Which GPR model was used for the comparison in Section 2.2, and on what settings?
8. How was reference 28 sourced? Its link ends in “utm_source=chatgpt.com” and it names CIGRE as the publisher.
9. Could you share the full set of 320 readings behind Tables 1 and 2?
10. The abstract says the AML “meets South Korea’s legal requirements for minimum depth and accuracy”. Was the AML ever put through the formal performance inspection for pipe locators, and did it pass?
11. Do either of you hold a role at the university centre that carries out that inspection?

Questions 10 and 11 were put to both authors on 23 September 2026.

I have also asked the journal to review the paper. A buyer shown this paper may rely on the abstract alone when deciding whether to trust a £7,250 handheld to judge where it is safe to dig.

DThe rest of the investigation