Chapter 12
Risks, safety and quality
Risk register, safety rules and quality assurance.
This chapter lists the main technical and organisational risks, the safety rules for laboratory and field work, and the quality measures that support the results.
12.1Risk register
Probability (P) and impact (I) are rated from 1 (low) to 3 (high); the score is their product. The register is reviewed monthly (WP0).
Table 12.1.Risk register, sorted by score.
| ID | Risk | P | I | Score | Mitigation | Owner |
|---|---|---|---|---|---|---|
| R2 | Coil nulling does not reach 10⁻³ at all tones | 2 | 3 | 6 | Adjustable overlap fixture; background subtraction per tone; gradiometer receiver pair | HW |
| R4 | Measured spectra do not match the model at M1 | 2 | 3 | 6 | Gate G1 with diagnosis time; independent reproduction; empirical calibration as fallback | PI |
| R7 | Dataset too small for the classification targets | 2 | 3 | 6 | Simulation; object-level augmentation; extension at gate G3 | DS |
| R15 | Grant not awarded, or awarded later or smaller than requested | 2 | 3 | 6 | Applications to several programmes in parallel (Chapter 11); a national grant funds WP1 and WP2 first; work packages can start in sequence as funds arrive | PI |
| R1 | Codec input noise or full-scale voltage worse than derived | 2 | 2 | 4 | The ×33 preamplifier is part of the design from the start; lower-noise op-amp option; measured in T1.3 | HW |
| R3 | Ground response larger than estimated, reducing range | 2 | 2 | 4 | Measured at two sites early (T2.6); spectral ground fitting; shallower test beds if needed | DS |
| R5 | RTK fix unavailable under trees or near buildings | 2 | 2 | 4 | Open test sites; camera and IMU bridging; wheeled cart with odometry (G2) | SW |
| R9 | Second AD9363 receive channel unavailable or phase unstable | 2 | 2 | 4 | Reference by switched loopback; radar scope reduced at gate G4 | HW |
| R10 | Procurement or customs delays | 2 | 2 | 4 | Long-lead items ordered in week 1; local alternatives; slack in WP3 and WP5 | PI |
| R8 | Copper and aluminium cannot be separated | 3 | 1 | 3 | Not required by any milestone; achieved separability is reported | PI |
| R11 | Loss of a team member | 1 | 3 | 3 | Two owners per work package; lab and decision logs; contingency for a contractor | PI |
| R14 | Safety incident during field work | 1 | 3 | 3 | Field rules in Chapter 12; only cleared land; safety briefing for every participant | HW |
| R6 | Magnetometer disturbed by the transmitter or by steel on site | 2 | 1 | 2 | Mounted at least 50 cm above the coil; heading from RTK track and camera odometry | SW |
| R12 | Host cannot sustain six-channel streaming | 1 | 2 | 2 | Both hosts tested in T1.2; reduced sample rate as fallback | SW |
| R13 | Radar emission outside permitted limits | 1 | 2 | 2 | Low power, shielded antennas facing the ground; regulations checked in T5.4 | HW |
12.2Safety
12.2.1Field work and ordnance
LODESTONE is a research instrument and is not a mine or ordnance detector. Several regions have known or suspected contamination with landmines and unexploded ordnance. The following rules apply without exception:
- Field work takes place only on prepared test beds or on land that is known to be free of ordnance, with the written permission of the owner.
- No work is carried out in areas marked or suspected as hazardous, or in their surroundings.
- If an object is found that might be ordnance, work stops, the location is marked from a distance and the relevant authorities are informed. The object is not touched.
- Excavation near archaeological or protected sites requires the permits required by national law; such sites are avoided.
12.2.2Electrical
- Transmit currents of up to 1 A per coil: connectors and wiring are rated accordingly and the drivers have current limiting.
- Lithium batteries are charged only with suitable chargers and transported in protective bags.
- No phantom power is connected to any coil input.
12.2.3Radio emissions
The radar sweeps 0.5 to 3 GHz and crosses bands used by mobile networks and satellite navigation. Ground-penetrating radar is a regulated device class in most countries. Transmit power is kept at the minimum needed, antennas are shielded and directed into the ground, and sweeps run only during measurements. The applicable national regulations are checked before field use (T5.4).
12.3Quality assurance
- Model checks. Every analytical model is verified against known limits before use, as done for the sphere solution in Chapter 3.
- Prediction before measurement. For each calibration measurement the expected result is written down before the measurement is taken.
- Calibration objects. Steel, copper, aluminium and stainless-steel spheres with known dimensions are measured at the start of each measurement day to detect drift.
- Independent checking. Results that feed a milestone are reproduced by a second team member from the raw data.
- Blind evaluation. Classification is only reported on objects that were not used in training, hidden by a person not involved in the analysis.
- Traceability. Every figure of measured data refers to raw files and a tagged software version.
12.4Open technical questions
- Actual input noise and full-scale voltage of the CS42448 board (T1.3).
- Whether 96 kHz sampling is reliable on the chosen host (T1.3).
- The ground-response factor at the test sites, which sets the detection range (T2.6).
- Achievable depth accuracy from peak width, which determines whether copper and aluminium can be distinguished (T3.7).
- Phase behaviour of the AD9363 synthesisers on the available Pluto unit and availability of the second channel (T5.1).
- RTK availability at the test sites (T3.2).