(Engineering Mathematics I & II — Foundation Automotive Technician Program)
This topic explains basic probability and statistical methods used in diagnostic decision-making, how to interpret measurement uncertainty, and how to apply simple quality checks in maintenance workflows — all with practical emphasis on low-cost tools and resource-constrained African workshop environments.
Learning objectives
After completing this topic, the learner will be able to:
- Distinguish between random and systematic errors and list common sources of measurement uncertainty in automotive work.
- Compute central tendency and dispersion (mean, median, mode, range, variance, standard deviation).
- Estimate measurement uncertainty and apply simple propagation rules.
- Use basic probability concepts to interpret diagnostic test results and make decisions under uncertainty.
- Apply simple quality-control checks (repeat measurements, basic control-chart thinking, acceptance by tolerance) in maintenance tasks.
1. Measurement errors and uncertainty — practical perspective
Types of error
- Systematic error: consistent bias (e.g., a mis-calibrated pressure gauge reads +0.5 bar). Tends to shift all measurements in one direction.
- Random error: variability from measurement to measurement (e.g., slight differences in reading a dial due to hand steadiness, temperature drift, or instrument noise).
- Resolution limits: smallest change the instrument can show (e.g., a multimeter that reads 0.1 V increments).
- Environmental and human factors: temperature, lighting, parallax when reading scales, inconsistent technique.
Practical steps to reduce uncertainty (low-cost)
- Zero and calibrate instruments using simple references (known battery, calibrated water column, or a reference weight).
- Warm up electronic instruments before use.
- Use the same instrument and method for repeated measurements to improve comparability.
- Take multiple measurements and record them.
- Read scales at eye level to avoid parallax.
- Keep records (date, instrument, operator, ambient conditions).
2. Basic descriptive statistics (for measurement sets)
Given a set of n measurements x1, x2, …, xn:
-
Mean (average): x̄ = (Σxi) / n
Use to estimate the central value. -
Median: middle value when data are ordered.
Less sensitive to outliers. -
Mode: most frequent value.
-
Range: max(x) − min(x).
Quick measure of spread. -
Variance (sample): s^2 = [Σ(xi − x̄)^2] / (n − 1)
-
Standard deviation (sample): s = sqrt(s^2)
Standard deviation quantifies the spread of repeated measurements. -
Standard error of the mean: s_x̄ = s / sqrt(n)
Use when estimating how well the mean represents the true value.
Example — Vernier caliper bore measurement (n = 5):
Measurements (mm): 79.92, 79.95, 79.94, 79.96, 79.93
x̄ = (79.92 + 79.95 + 79.94 + 79.96 + 79.93) / 5 = 79.94 mm
Compute s and s_x̄ to evaluate consistency.
3. Reporting measurement uncertainty
- Report measurements as: measured value ± uncertainty (coverage). Example: 79.94 mm ± 0.03 mm (k = 2, ~95% confidence).
- For practical workshop use, k = 2 is commonly used to represent an expanded uncertainty (~95% coverage) when no formal laboratory calibration is possible.
Practical rule-of-thumb:
- If s_x̄ is the standard error, expanded uncertainty U ≈ k·s_x̄ with k = 2 for most diagnostic decisions.
4. Error analysis and uncertainty propagation (simple rules)
When combining measurements to compute derived quantities, propagate uncertainties.
Let u(x) denote the absolute standard uncertainty in x.
-
Addition / subtraction:
z = x ± y
u(z) = sqrt[u(x)^2 + u(y)^2] -
Multiplication / division:
z = x * y or z = x / y
relative uncertainty: u_rel(z) = sqrt[ u_rel(x)^2 + u_rel(y)^2 ]
where u_rel(x) = u(x)/x -
Power:
z = x^a
u_rel(z) = |a| · u_rel(x)
Example — Fuel flow calculation:
You measure fuel mass m = 12.0 ± 0.2 kg and time t = 3600 ± 5 s, compute mass flow ṁ = m / t.
Relative uncertainties:
u_rel(m) = 0.2 / 12.0 = 0.0167
u_rel(t) = 5 / 3600 = 0.00139
u_rel(ṁ) = sqrt(0.0167^2 + 0.00139^2) ≈ 0.0168
So ṁ relative uncertainty ≈ 1.68%.
5. Percent error and accuracy
- Absolute error = measured − true (or reference).
- Percent error = (measured − true) / true × 100%
Example:
Reference cylinder bore = 80.00 mm, measured x̄ = 79.94 mm → percent error = (−0.06 / 80.00) × 100% = −0.075%.
6. Basic probability concepts for diagnostics
- Probability: measure of how likely an event is to occur (0 to 1).
- Events: e.g., "injector is leaking", "pressure test positive".
- Independent events: outcome of one does not affect another.
- Conditional probability: probability of event A given event B (P(A|B)).
Bayes’ theorem (practical form):
P(condition | positive test) = [P(positive | condition) · P(condition)] / P(positive)
This is useful when interpreting test results: the probability a component is faulty given a positive test depends on the prior probability (prevalence) of that fault and the test’s accuracy.
Example — Simple Bayes calculation:
- Prevalence (prior) of injector fault on a fleet: P(Fault) = 0.10 (10%).
- Test sensitivity: P(Positive | Fault) = 0.9 (90% chance test is positive if faulty).
- Test false positive rate: P(Positive | No Fault) = 0.1 (10%).
P(Positive) = 0.9·0.1 + 0.1·0.9 = 0.09 + 0.09 = 0.18
P(Fault | Positive) = 0.09 / 0.18 = 0.5
Interpretation: given a positive test result, there is a 50% chance the injector is truly faulty — so further confirmation (additional tests) is advisable.
7. Diagnostic decision-making: sensitivity, specificity, and thresholds
- Sensitivity: probability test detects the fault when it exists (true positive rate).
- Specificity: probability test is negative when no fault exists (true negative rate).
- False positives and negatives lead to unnecessary repairs or missed faults.
Decision rules in workshops (practical):
- If measurement ± uncertainty crosses the acceptance limit, classify as "borderline" and retest or use secondary method.
- If measurement is clearly outside tolerance beyond expanded uncertainty (e.g., measurement = 79.0 mm ± 0.05 mm and tolerance lower limit is 79.5 mm), take corrective action.
- When tests are not definitive (e.g., Bayes result ≈ 50%), perform additional independent tests.
Example — Battery voltage decision:
- Acceptance threshold for fully charged 12 V battery: 12.6 V.
- Multimeter uncertainty: ±0.1 V.
- Measured voltage: 12.55 V ± 0.1 V.
Because 12.55 ± 0.1 includes 12.6 V, the result is borderline. Recommended action: charge battery and re-test, or use a load test to confirm.
8. Simple quality checks and control-chart thinking
You can implement basic statistical quality control without expensive software:
- Repeatability check: take 3–5 measurements and compute mean and standard deviation. Use standard deviation to judge consistency.
- Simple control rule: if a measurement deviates more than 2 standard deviations from the process mean, investigate.
- Trend detection: plot sequential measurements (simple paper chart). A run of 6–8 points trending in one direction indicates a developing issue even within control limits.
- Acceptance by tolerance: if x̄ ± U remains within component tolerance, accept; if it lies outside, reject; if it overlaps, classify as borderline.
Example — Brake disc thickness:
Target thickness = 22.0 mm, wear limit = 21.0 mm. Measurements across the disc ring: 21.95, 21.90, 21.92, 21.98, 21.94 → mean 21.938 mm, s ≈ 0.03 mm, s_x̄ ≈ 0.013 mm. Even considering uncertainty, mean > 21.0 mm so disc is acceptable. If one location measured 20.98 mm (less than limit), mark for replacement.
9. Practical worked examples
Example 1 — Cylinder bore (acceptance vs specification)
- Specification: bore = 80.00 ± 0.10 mm (acceptable if between 79.90 and 80.10 mm).
- Measurements (mm): 79.92, 79.95, 79.94, 79.96, 79.93
- Mean = 79.94 mm, s ≈ 0.014 mm, s_x̄ ≈ 0.0063 mm
- Expanded uncertainty (k=2): U ≈ 2·0.0063 ≈ 0.0126 mm
- Report: 79.94 ± 0.013 mm — well within tolerance. Action: accept.
Example 2 — Pressure test for a leak (Bayesian thinking)
- Prior probability of leak in system: 0.05 (5%).
- Test: simple pressure drop test that is positive with sensitivity 0.8 and false positive rate 0.1.
- P(Positive) = 0.80.05 + 0.10.95 = 0.04 + 0.095 = 0.135
- P(Leak | Positive) = 0.04 / 0.135 ≈ 0.296 (≈30%)
- Interpretation: a positive pressure drop test increases the likelihood of leak (from 5% to ~30%) but is not conclusive; inspect further.
Example 3 — Propagation: calculated torque from force and lever arm
- Measured force F = 200.0 ± 1.0 N; lever arm L = 0.300 ± 0.003 m.
- Torque T = F × L = 200.0 × 0.300 = 60.0 Nm
- Relative uncertainties: u_rel(F) = 1/200 = 0.005; u_rel(L) = 0.003/0.300 = 0.01
- u_rel(T) = sqrt(0.005^2 + 0.01^2) ≈ 0.01118
- Absolute uncertainty u(T) = 0.01118 × 60.0 ≈ 0.67 Nm
- Report: T = 60.0 ± 0.7 Nm (k = 1). For k = 2, ±1.4 Nm.
10. Practical checklist for measurement and diagnostic workflows
- Identify the critical measurement and its tolerance/specification.
- Select the simplest, lowest-cost suitable instrument.
- Verify instrument zero/calibration against a simple reference.
- Record operator, instrument, ambient conditions.
- Take at least 3 independent measurements; more when variability is high.
- Compute mean and standard deviation; calculate standard error.
- Report value with uncertainty and compare to specification using the conservative rule:
- If (mean − U) ≥ lower limit and (mean + U) ≤ upper limit → Accept.
- If (mean + U) < lower limit or (mean − U) > upper limit → Reject.
- Otherwise → Borderline: repeat measurement or perform complementary test.
- If diagnostic test is not definitive, consider prior probability and perform secondary tests before major repair.
11. Record keeping and communicating uncertainty
- Always log raw measurements, computed statistics, instruments used, and any adjustments made.
- Communicate uncertainty plainly: “Measured = 12.55 V ± 0.10 V (multimeter uncertainty). Borderline with recommended additional testing.”
- For fleet management, keep simple charts of repeated measurements to identify trends and schedule preventive maintenance.
12. Quick reference: common formulas
- Mean: x̄ = (Σxi) / n
- Sample standard deviation: s = sqrt[Σ(xi − x̄)^2 / (n − 1)]
- Standard error: s_x̄ = s / sqrt(n)
- Addition/subtraction uncertainty: u(z) = sqrt[u(x)^2 + u(y)^2]
- Multiplication/division relative uncertainty: u_rel(z) = sqrt[u_rel(x)^2 + u_rel(y)^2]
- Percent error = (measured − reference) / reference × 100%
- Expanded uncertainty (approx.): U ≈ k·s_x̄ (k = 2 for ~95% confidence)
Closing guidance for resource-constrained contexts
- Statistical thinking and error analysis do not require expensive tools. Repeated consistent measurements, simple arithmetic, and conservative decision rules dramatically reduce costly misdiagnoses.
- Use low-cost references (known weights, reference voltages, or master parts) to check instruments regularly.
- When uncertain, prefer additional low-cost confirmatory tests before significant repairs.
- Teach colleagues these simple methods so diagnostic quality improves across the workshop.
End of topic.