Fault diagnosis

Why this matters

“The drone leans right, so replace the right motor” is a guess that wastes money and time, and if it is wrong the symptom returns on the next flight. A good technician works like a doctor: listen to the symptoms, form several hypotheses, then choose tests that tell the hypotheses apart before deciding on a treatment. This module practises that way of thinking with real flight-controller logs.

The fault tree

A fault tree puts the unwanted symptom at the top and branches down into possible causes. An OR gate means any one of the causes below can produce the event above. The tree helps you not to forget whole groups of causes you had not thought of.

The diagnosis loop

Three key principles:

• Separate observation from interpretation. Write “at 10:02:15 voltage fell from 23.6 to 22.7 V while current rose”, not “the battery is worn”
• Choose tests whose results differ between hypotheses. If a test gives the same result whichever hypothesis is true, it cannot decide between them
• A log is evidence of what the system recorded, not the whole truth. An abnormal value may come from a sensor, a cable or a communication link, not necessarily from the part the value describes

The ArduPilot documentation explains how to download and read logs with Mission Planner. Commonly used messages include BAT (battery), RCOU (outputs to the motors), VIBE (vibration) and ATT (attitude).

Evidence from voltage and current

A battery has internal resistance, like a narrow water pipe: the more current you draw, the more the terminal voltage sags. A worn battery or a loose connector raises the total resistance. It can be estimated from two steady segments of the same flight.

v_hover, i_hover = 23.62, 19.8   # steady hover segment
v_climb, i_climb = 22.71, 46.5   # steady climb segment a few seconds later
r_pack = (v_hover - v_climb) / (i_climb - i_hover)
print(f"pack resistance about {r_pack * 1000:.1f} mΩ (register value when new: 17.0 mΩ)")

pack resistance about 34.1 mΩ (register value when new: 17.0 mΩ)

Evidence from motor commands

In a steady hover, all motors should receive similar commands. If one motor is consistently commanded higher than the others, the controller is making it work harder to hold attitude.

rcou = {1: 1542, 2: 1538, 3: 1631, 4: 1529}   # mean PWM (µs) per motor during a steady hover
mean = sum(rcou.values()) / len(rcou)
for motor, pwm in rcou.items():
    print(f"motor {motor}: {pwm} µs ({pwm - mean:+.0f} from mean)")

motor 1: 1542 µs (-18 from mean)
motor 2: 1538 µs (-22 from mean)
motor 3: 1631 µs (+71 from mean)
motor 4: 1529 µs (-31 from mean)

Missing data

Logs may have gaps, for example when the telemetry link drops. A gap is not a zero. Filling missing values with 0 creates a “voltage drop to zero” that never happened.

import pandas as pd

log = pd.DataFrame({"time_s": [0, 1, 2, 3, 4],
                    "volt": [23.1, 23.0, None, None, 22.9],
                    "link": ["ok", "ok", "lost", "lost", "ok"]})
print("min voltage if gaps are read as 0:", log["volt"].fillna(0).min())
print("min voltage from real samples:   ", log["volt"].min())
gaps = log["volt"].isna()
print("missing samples:", int(gaps.sum()), "| all during link loss:", bool((log.loc[gaps, "link"] == "lost").all()))

min voltage if gaps are read as 0: 0.0
min voltage from real samples:    22.9
missing samples: 2 | all during link loss: True

The missing samples coincide exactly with the link loss, which supports a communication problem rather than a voltage drop, but it does not prove the physical cause of the link loss. If an onboard (DataFlash) log exists for the same period, open it, because it does not depend on the radio link.

Module lab

Common mistakes

Summary

• A fault tree helps list all causes before starting repairs
• Diagnose with several hypotheses, and choose tests whose results differ between them
• Internal resistance is estimated as voltage drop divided by current rise, and includes connectors and wiring
• Imbalanced motor commands show which motor works hard, but not yet why
• Missing data must stay as gaps, and conclusions must state their confidence

Check your understanding

1. Voltage falls from 24.0 to 23.4 V as current rises from 10 to 40 A. What is the approximate total resistance?
2. Four motors receive 1500, 1500, 1500 and 1580 µs. How far is the last motor from the mean?
3. What does an OR gate in a fault tree mean?
4. Why should missing voltage data not be replaced with zeros?
5. Total resistance is abnormally high. Which test separates a battery cause from an aircraft connector cause?