Here is what you are about to do.
You are not being asked to know electronics from memory. You will be given a system model and its reference relationships, then use them to decide what should be true, what to measure, what the evidence means, and whether a repair actually restored the system.
Given an unfamiliar but bounded system, externalize the model, preserve rival explanations, buy discriminating evidence, locate the first abnormal boundary, make a bounded change, and prove recovery.
A temperature logger says 57.5 °C. A trusted reference says 24.0 °C.
The display is wrong, but that does not tell you which part is wrong. You will trace a small signal chain from physical temperature to sensor voltage, buffer output, ADC conversion, software, and display.
- The equations and released reference values stay available.
- You get only three diagnostic measurements before intervention.
- You must predict before revealing each result.
- A cosmetic fix will not pass independent verification.
- At the end, you will transfer the same method to a different Black Knight telemetry fault.
Reference access is not a failure of competence.
Start with the system you think you have.
A low-voltage environmental logger is being checked against a calibrated 24.0 °C reference block. Its display reports 57.5 °C and raises a high-temperature warning. You are authorized to use the designated 0–5 V test points and to swap the plug-in ADC reference module. You are not authorized to redesign the circuit or work on hazardous energized equipment.
Locate the first abnormal state, choose a discriminating test, make one bounded intervention, and verify recovery against predeclared expectations.
Diagnosis is a comparison problem.
- Model
- Name the states and transformations that should connect cause to indication.
- Expected state
- Derive what should be observable at useful boundaries before measuring.
- Discrepancy
- The first point where observed state no longer matches expected state bounds the fault region.
- Discriminating test
- Prefer a measurement that makes plausible hypotheses predict different results.
- Verification
- Declare what must become true before the intervention. A green display is not an independent check.
Not equation recall.
- Recognize which relationship applies.
- Use the supplied relationship correctly.
- Check scale and units.
- Interpret whether the result fits the proposed mechanism.
VS = 0.50 V + (0.010 V/°C)TT is temperature in °C. The constant and slope are supplied by the sensor model.
VB = 2VSThe buffer's released gain is 2.
N ≈ 4095 × VIN / VREFN is ADC count. Lower VREF raises the count for the same input.
(observed − expected) / expectedUseful for comparing the direction and scale of a discrepancy.
At 20 °C: VS = 0.50 + 0.010×20 = 0.70 V. The buffer gives 1.40 V. With VREF = 2.50 V, the ADC predicts about 4095×1.40/2.50 = 2293 counts.
- Substitute T = 24.0 °C into the sensor law to get the expected sensor voltage.
- Use that voltage as the input to the released ×2 buffer.
- Use the resulting buffer voltage as VIN in the 12-bit ADC relationship with VREF = 2.50 V.
- Run the same ADC relationship separately for the independent 1.000 V calibration input.
- Keep units attached. The arithmetic supports the model; the later diagnostic task is deciding what a mismatch means.
Use the supplied relationships. Arithmetic is not the point; the point is knowing what should connect the numbers. Values within normal rounding are accepted.
Make rival explanations earn different predictions.
The board keeps the reasoning outside your head. Choose two live rivals, then choose a measurement whose expected result differs between them. If both explanations predict the same result, the test cannot discriminate between them.
A serious hypothesis predicts what should be observed before the observation arrives.
System reference · 24.0 °C trusted reference · display 57.5 °C · nominal VREF 2.50 VOpen model + derivation
| Hypothesis | What it predicts | High-value boundary | Evidence that weakens it |
|---|---|---|---|
| H1 — Actual temperature high | Local reference, TP-S and TP-B should all be high. | Independent local temperature | The calibrated reference at the sensor site remains 24.0 °C. |
| H2 — Sensor biased high | TP-S high; TP-B high; VREF normal; calibration channel normal. | TP-S | TP-S matches the sensor law at 24.0 °C. |
| H3 — Buffer gain fault | TP-S normal; TP-B departs from 2×TP-S; VREF and calibration channel normal. | TP-S → TP-B | TP-B remains 2×TP-S. |
| H4 — ADC reference low | TP-S and TP-B normal; TP-REF low; calibration counts inflated. | TP-REF + independent calibration channel | VREF and calibration channel both match healthy expectation. |
| H5 — Conversion coefficients wrong | Analog path, VREF and raw calibration counts normal; released configuration does not match. | Configuration + raw count | Released coefficients match and raw counts are already wrong. |
A list of possible causes is not yet reasoning.
The useful step is converting each cause into an observable prediction, then buying the measurement that separates them.
Buy information, then quantify the discrepancy.
You have three diagnostic measurements before intervention. The reference relationships remain available. Before each result appears, identify the relationship you are using and state what the measurement is supposed to discriminate.
Three evidence requests. A successful path must establish an upstream analog state, measure VREF, and use the independent calibration channel.
System reference · 24.0 °C trusted reference · display 57.5 °C · nominal VREF 2.50 VOpen model + derivation
Relationship reference
V_S = 0.50 + 0.010T — sensor expectation.V_B = 2V_S — buffer expectation.N ≈ 4095 V_IN / V_REF — ADC scaling.(observed−expected)/expected — relative deviation.If you spend the budget without establishing both the causal scale check and an upstream normal state, do not intervene. Restart with a better test strategy.
Name the first state that broke the model.
A diagnosis should explain every relevant observation with the fewest unsupported assumptions. It should also say what the current evidence does not establish.
Diagnosis + first abnormal boundary + surviving uncertainty.
System reference · 24.0 °C trusted reference · display 57.5 °C · nominal VREF 2.50 VOpen model + derivation
The downstream symptom can be honest.
With a low ADC reference, the converter can faithfully report a high code even though the sensor and buffer voltages are normal. The display is wrong about temperature because the measurement chain's reference assumption is wrong.
Do not infer a component failure from a single abnormal reading if the reference used to create that reading has not been checked.
Change the smallest thing your diagnosis actually supports.
You are authorized to swap the plug-in 2.50 V reference module. Before acting, choose the intervention and define the evidence that would count as recovery.
Diagnosis → authorized action → predeclared verification.
System reference · 24.0 °C trusted reference · display 57.5 °C · nominal VREF 2.50 VOpen model + derivation
Select the checks that must pass. Choose at least three; the independent reference and ADC reference path must be represented.
Calibration theater
Changing a software offset can make one number look correct while leaving the ADC reference wrong. A cosmetic match is not closure if an independent calibration channel still fails.
The intervention is not the result.
Compare the post-action state with the criteria you committed before the change. Then classify the outcome as pass, fail or inconclusive.
Expected → observed → pass / fail / inconclusive. Diagnose mismatch before declaring recovery.
System reference · 24.0 °C trusted reference · display 57.5 °C · nominal VREF 2.50 VOpen model + derivation
| Check | Observed | Meaning |
|---|
The 1.000 V calibration input is valuable because it bypasses the temperature sensor and buffer. If it recovers at the same time as the temperature channel, the diagnosis gains support from a separate path.
Same method. Different physics.
A coolant loop's central display reports 0.91 bar while a local mechanical reference reads 2.10 bar. The training problem is now digital telemetry rather than an analog ADC reference. The method must survive the surface change.
Find the first abnormal boundary without assuming the previous fault repeats.
Open reference shelf
Δt = t_current − t_packet — age of a reported state.(observed−expected)/expected — relative deviation when scale matters.The previous fault was an analog reference error; this one is a stale digital state. The test is whether the same causal grammar survives when the failed component changes.
Keep the causal record.
The useful record preserves the model, rival mechanisms, discriminating measurements, intervention, and independent verification so another operator can reconstruct why the diagnosis won.
Cross-context diagnostic reasoning only; physical bench work remains external.
Keep the method current by diagnosing unfamiliar but bounded systems. If practice becomes “replace the likely part until it works,” the capability has degraded even if the vocabulary remains familiar.