Radiation Detector Dead-Time Count Loss Case Study

\displaystyle n=\frac{m}{1-m\tau}

\dot{H}_{display}=C(m-b)

\dot{H}_{display}=0.010(16{,}500-120)

\dot{H}_{display}=163.8\ \mu\text{Sv/h}

\displaystyle U_{display}=\frac{163.8}{200}=0.82

m\tau=16{,}500(22\times 10^{-6})=0.363

1-m\tau=0.637

\displaystyle n=\frac{16{,}500}{0.637}=25{,}903\ \text{s}^{-1}

n_b\approx 25{,}903-120=25{,}783\ \text{s}^{-1}

\dot{H}_{corr}=0.010(25{,}783)=257.8\ \mu\text{Sv/h}

\displaystyle U_{corr}=\frac{257.8}{200}=1.29

\displaystyle \frac{257.8-163.8}{257.8}=0.365$$ So the displayed value understated the corrected dose-rate estimate by about: $$36.5\%$$ ### Engineering Comment The result is not a small calibration trim. A 36% difference changes the engineering decision. A monitor can pass a simple threshold screen while failing the physical validity of the measurement chain. ## Step 4: Count-Rate Linearity Check The team repeated the test with controlled source output fractions and recorded observed count rate. | Source setting | Observed count rate $m$ | $m\tau$ | Corrected count rate $n=m/(1-m\tau)$ | Corrected rate divided by source setting | | ---: | ---: | ---: | ---: | ---: | | 0.25 | $6{,}300\ \text{s}^{-1}$ | 0.139 | $7{,}317\ \text{s}^{-1}$ | $29{,}268\ \text{s}^{-1}$ | | 0.50 | $11{,}200\ \text{s}^{-1}$ | 0.246 | $14{,}854\ \text{s}^{-1}$ | $29{,}708\ \text{s}^{-1}$ | | 1.00 | $16{,}500\ \text{s}^{-1}$ | 0.363 | $25{,}903\ \text{s}^{-1}$ | $25{,}903\ \text{s}^{-1}$ | The 0.25 and 0.50 settings agree after correction. The 1.00 setting remains low compared with the lower-rate slope: $$\frac{25{,}903}{29{,}500}\approx 0.88$$ This indicates that the simple dead-time correction improved the estimate but did not fully recover linearity at the highest source setting. ### Engineering Comment This is the decisive evidence. If the only issue were non-paralyzable dead time with a stable $\tau$, corrected rates would scale cleanly with source setting. The high-rate residual error points to additional pulse pileup, discriminator threshold shift, analog saturation, or firmware rejection. That makes the high-rate correction unsuitable as a release basis by itself. ## Step 5: Poisson Counting Uncertainty For the 10 second integration window, the observed counts were: $$N=mT=16{,}500(10)=165{,}000$$ Poisson standard uncertainty in observed counts is: $$\sigma_N=\sqrt{165{,}000}=406$$ Observed count-rate standard uncertainty: $$u_m=\frac{406}{10}=40.6\ \text{s}^{-1}$$ For: $$n=\frac{m}{1-m\tau}$$ the sensitivity to observed count rate is: $$\frac{\partial n}{\partial m}=\frac{1}{(1-m\tau)^2}$$ At $m\tau=0.363$: $$\frac{\partial n}{\partial m}=\frac{1}{0.637^2}=2.46$$ So: $$u_n\approx 2.46(40.6)=100\ \text{s}^{-1}$$ Dose-rate uncertainty from counting statistics alone is: $$u_{\dot{H},count}=0.010(100)=1.0\ \mu\text{Sv/h}$$ Counting noise is not the dominant uncertainty. Dead-time knowledge and calibration are. ## Step 6: Dead-Time and Calibration Uncertainty The sensitivity of corrected count rate to dead time is: $$\frac{\partial n}{\partial \tau}=\frac{m^2}{(1-m\tau)^2}$$ Using $m=16{,}500\ \text{s}^{-1}$ and $1-m\tau=0.637$: $$\frac{\partial n}{\partial \tau}= \frac{16{,}500^2}{0.637^2}=6.72\times 10^8\ \text{s}^{-2}$$ For $u_\tau=2\ \mu\text{s}$: $$u_{n,\tau}=6.72\times 10^8(2\times 10^{-6})=1{,}344\ \text{s}^{-1}$$ Dose-rate contribution: $$u_{\dot{H},\tau}=0.010(1{,}344)=13.4\ \mu\text{Sv/h}$$ Calibration standard uncertainty: $$u_{\dot{H},cal}=0.08(257.8)=20.6\ \mu\text{Sv/h}$$ Combined standard uncertainty, treating the contributors as independent: $$u_c=\sqrt{1.0^2+13.4^2+20.6^2}=24.6\ \mu\text{Sv/h}$$ The corrected estimate can be stated as approximately: $$\dot{H}_{corr}=258\pm 25\ \mu\text{Sv/h}\quad(k=1)$$ Even one standard uncertainty below the estimate: $$258-25=233\ \mu\text{Sv/h}$$ remains above the $200\ \mu\text{Sv/h}$ review threshold. ### Engineering Comment The conclusion does not depend on a fragile arithmetic detail. The corrected value is high enough that ordinary counting uncertainty cannot explain it away. The dominant issue is not random noise; it is measurement-chain validity. ## Failure Mode Diagnosis | Evidence | Interpretation | | --- | --- | | Displayed dose rate below threshold | Raw counts alone appeared acceptable. | | $m\tau=0.363$ | Detector was outside a comfortable low-loss operating region. | | Corrected dose rate above threshold | Count loss changed the release decision. | | High source setting failed corrected linearity | Dead-time model alone was insufficient at the peak condition. | | Independent survey meter read higher | The installed monitor was not a reliable sole release instrument. | | No saturation or live-time alarm | Software did not warn operators that the measurement was outside its valid range. | The most likely combined failure mode was high count-rate operation with pulse pileup and dead-time loss. The detector did not fail silent; it failed plausibly. ## Corrective Action The release path used four changes. 1. Re-range the detector channel so normal operation keeps $m\tau<0.10$ at the alarm threshold. 2. Add live-time correction and display both observed and corrected count rates in service data. 3. Add a saturation alarm when $m\tau>0.20$ or when source-step linearity exceeds the allowed residual. 4. Require independent survey confirmation after source, shielding, detector gain, pulse-shaping, or firmware changes. The $m\tau<0.10$ screening limit gives: $$m_{max}=\frac{0.10}{22\times 10^{-6}}=4{,}545\ \text{s}^{-1}$$ For the old detector chain, routine operation near $16{,}500\ \text{s}^{-1}$ was not acceptable. The team either had to reduce the event rate at that detector, use a faster chain, change geometry, add attenuation, or choose a monitor designed for the higher rate. ## Post-Correction Validation After re-ranging, the monitor used a faster processing chain and revised alarm firmware. Validation used reference fields and an independent calibrated survey instrument. | Check | Acceptance criterion | Result | | --- | --- | --- | | low-rate calibration | within $\pm 6\%$ of reference | passed | | source-step linearity | corrected slope residual below $5\%$ | passed | | alarm threshold test | trip before corrected dose rate exceeds threshold | passed | | saturation flag | active before invalid count-rate region | passed | | background subtraction | stable within documented uncertainty | passed | | independent survey confirmation | agreement within combined uncertainty | passed | The revised release report included: - source setting and geometry; - detector dead-time value and method used to estimate it; - count-rate range over which calibration is valid; - observed count rate, corrected count rate, and live-time fraction; - independent survey comparison; - uncertainty budget and decision rule; - alarm and saturation tests; - configuration control for gain, discriminator, shaping time, and firmware. ## Engineering Lessons 1. A detector reading can look stable while being wrong at high event rate. 2. Radiation-dose arithmetic is not enough; detector linearity and count-rate validity must be checked. 3. Dead-time correction is a diagnostic tool, not an automatic permission to operate beyond the validated range. 4. Counting statistics may be much smaller than calibration, geometry, dead-time, and model-form uncertainty. 5. Alarm systems should fail visibly when the measurement chain enters a region where the displayed value is no longer trustworthy. The transferable lesson is that measurement validity is part of system safety. A radiation monitor, vibration channel, photodiode, thermocouple, or pressure gauge is only useful inside the range where its physics, electronics, calibration, and software decision rule have been validated.