Research Design and Engineering Methodology

Introduction

A major project in electrical engineering is not merely a device, simulation, software package, or written report. It is a defensible investigation that connects an engineering problem to evidence, analysis, design decisions, and conclusions. Research design is the plan that explains how evidence will be obtained and interpreted. Engineering methodology is the disciplined sequence of activities used to transform a need or question into a verified result.

A strong project proposal therefore answers six questions:

1. What problem is being addressed?
2. Why is the problem important?
3. What is already known, and what gap remains?
4. What method will produce credible evidence?
5. How will performance be measured and compared?
6. Can another competent engineer reproduce the work?

Electrical engineering projects commonly combine theory, experimentation, simulation, embedded implementation, data analysis, and prototyping. The appropriate method depends on the research question, available resources, safety constraints, required precision, and the type of claim being made. A simulation cannot by itself establish that a physical converter is safe; a prototype test cannot by itself explain every causal mechanism; and a literature review cannot substitute for measurement when the claim concerns actual hardware performance.

Learning Objectives

By the end of this chapter, a student should be able to:

• distinguish research paradigms and select a suitable methodology;
• formulate a precise engineering problem, aim, objectives, and research questions;
• identify variables, assumptions, controls, constraints, and sources of error;
• design quantitative, qualitative, experimental, simulation, prototyping, and mixed-method studies;
• select defensible samples, test cases, metrics, and comparison baselines;
• plan reproducible procedures and appropriate data analysis;
• address validity, reliability, safety, ethics, intellectual property, and data governance;
• prepare a coherent major-project proposal.

1. From Engineering Problem to Researchable Question

An engineering problem usually begins as a practical need: reduce inverter losses, improve battery-state estimation, detect faults earlier, lower electromagnetic interference, or increase the reliability of a motor drive. Such statements are valuable but too broad for a defensible investigation. They must be converted into a bounded research problem.

A useful problem statement identifies the present condition, deficiency, context, consequence, and proposed direction. For example: “A low-cost photovoltaic inverter used in a small laboratory microgrid exhibits excessive output-voltage distortion during rapid irradiance changes. Existing control methods improve steady-state performance but are computationally demanding for the selected microcontroller. This project will evaluate whether a computationally efficient predictive control strategy can reduce distortion without exceeding the processor’s execution-time limit.”

The aim expresses the overall intended achievement. Objectives divide the aim into observable tasks. A good objective begins with a measurable verb such as design, model, implement, compare, quantify, validate, or evaluate. “Understand inverter control” is weak; “compare the total harmonic distortion of two control algorithms under three defined load conditions” is testable.

Research questions should specify the population or system, intervention or condition, outcome, and comparison where appropriate. A hypothesis is a testable prediction, not a conclusion. For example:

• Research question: How does switching frequency affect converter efficiency at a fixed output power?
• Null hypothesis: Mean efficiency does not differ across the selected switching-frequency conditions.
• Alternative hypothesis: At least one switching-frequency condition produces a different mean efficiency.

A project may instead use design questions, such as “Can a sensor-fusion algorithm estimate battery state of charge within ±3% over the specified drive cycle?” Design questions are answered through requirements, implementation, verification, and validation rather than statistical significance alone.

2. Research Paradigms and Method Selection

A paradigm is a broad view of what counts as knowledge and how it should be justified. Most undergraduate electrical-engineering projects are pragmatic: the method is selected because it is suitable for answering the question and producing a useful, verifiable engineering outcome.

Quantitative research

Quantitative research represents observations numerically and analyses relationships, differences, trends, or uncertainty. It is appropriate for questions involving efficiency, temperature, voltage error, accuracy, failure rate, latency, energy consumption, or statistical association. Its strengths include explicit metrics, repeatability, and comparison. Its limitations include dependence on measurement quality and the risk of reducing a complex system to incomplete indicators.

Qualitative research

Qualitative research examines meanings, experiences, processes, or expert judgments using interviews, observations, documents, or open-ended responses. It may be appropriate when investigating maintenance practices, operator interaction with a protection system, usability of an engineering interface, or barriers to technology adoption. A qualitative study still requires a systematic sampling and analysis procedure; it is not informal opinion gathering.

Experimental research

An experiment deliberately changes one or more independent variables and observes the effect on dependent variables while controlling relevant conditions. Randomisation, replication, blocking, and a baseline or control condition strengthen causal inference. In engineering, a fully controlled experiment may be performed on a test bench, environmental chamber, hardware-in-the-loop platform, or laboratory prototype.

Simulation and computational research

Simulation uses a mathematical, circuit, finite-element, system-level, or numerical model to predict behaviour. It is efficient for exploring parameter ranges and unsafe or expensive conditions. A simulation study must document equations or software models, solver settings, initial conditions, parameter sources, tolerances, and validation against analytical or measured results. A visually convincing plot is not evidence of model validity unless the model has been checked.

Prototyping and design research

Prototyping develops and evaluates an artefact. The process may include requirements analysis, architecture, detailed design, implementation, verification, validation, and iteration. A prototype is not automatically a validated product. Its claims must be limited to the tested operating range, components, environmental conditions, and maturity level.

Mixed methods

Mixed methods deliberately combine quantitative and qualitative evidence. For example, a smart-meter project may measure classification accuracy while interviewing users about alarm interpretability. Integration must be planned: qualitative findings may explain quantitative results, identify variables, or assess practical acceptability. Merely placing a survey beside an experiment does not create a coherent mixed-method design.

Method-selection framework

Select a method by matching the claim to the evidence required.

3. Conceptual Model, Variables, and Assumptions

A conceptual model states how the project expects inputs, mechanisms, and outputs to be related. It can be expressed in a block diagram, a set of equations, or a written causal chain. The model prevents uncontrolled expansion of the project and guides measurement selection.

The independent variable is deliberately changed or classified as a predictor. The dependent variable is measured as an outcome. Controlled variables are held constant or balanced. Confounding variables affect both the apparent cause and outcome, creating a misleading relationship. A mediator explains part of the mechanism between cause and outcome, while a moderator changes the strength or direction of that relationship.

For a DC–DC converter, switching frequency may be the independent variable; efficiency and output-voltage ripple may be dependent variables; input voltage, output power, ambient temperature, load type, and measurement bandwidth may be controlled variables. Component temperature may be a mediator because frequency affects losses, losses affect temperature, and temperature affects efficiency.

Assumptions are conditions accepted for analysis. Examples include steady-state operation, negligible probe loading, known sensor bias, constant ambient temperature, or independent measurement errors. Every important assumption should be justified and tested where possible. An assumption that is false can invalidate a conclusion even when calculations are correct.

Measurement definitions

Operational definitions state exactly how a construct is measured. “Power quality” might be defined through voltage total harmonic distortion, frequency deviation, flicker, or unbalance. “Algorithm performance” might include accuracy, precision, recall, execution time, memory usage, and energy per inference. Definitions should specify the instrument, sampling rate, time window, preprocessing, units, and aggregation method.

For a periodic voltage waveform with fundamental RMS component V₁ and harmonic RMS components V₂, V₃, and so on, voltage total harmonic distortion is commonly defined as:

Here Vₖ is the RMS magnitude of the kth harmonic, n is the highest included harmonic, and V₁ must be non-zero. The measurement bandwidth, windowing method, harmonic range, and instrumentation must be reported. The expression is not interchangeable with a waveform’s total RMS value.

For an electrical converter, efficiency is:

where P_out and P_in are average output and input powers in watts, and P_in must be positive for normal operation. The method used to calculate power—instantaneous voltage-current averaging or separately measured DC power—must be stated, especially for nonsinusoidal waveforms.

4. Experimental Design

A defensible experiment begins with a test matrix. The matrix lists factors, levels, combinations, repetitions, and required measurements. A single reading at each condition cannot reveal repeatability or random variation. Replication means repeating measurements under nominally identical conditions; repeated measures on the same specimen are not necessarily independent samples.

A simple one-factor experiment may compare three switching frequencies at four output-power levels, with five repeated runs per combination. If thermal history influences results, the run order should be randomised or balanced. If tests must be performed on different days, day can be treated as a blocking factor. If a complete factorial design is affordable, it can reveal interactions; for factors A and B, an interaction exists when the effect of A changes across levels of B.

Controls may be:

• a conventional algorithm against which a new algorithm is compared;
• a no-intervention or baseline condition;
• a reference instrument or calibrated standard;
• a negative control that should not produce the expected response;
• a positive control known to produce a response.

The control must be technically comparable. Comparing a new controller on one hardware platform with an old controller on another may confound algorithm and hardware effects.

Sampling and sample size

The target population is the complete set to which the claim is intended to apply. The accessible population is the portion available for study. A sample is selected from that population. In electrical engineering, samples may be devices, batteries, motors, datasets, operating points, users, or environmental conditions.

Probability sampling supports statistical generalisation when a sampling frame exists. Simple random, stratified, systematic, and cluster sampling each have different implications. Convenience sampling may be acceptable for an exploratory prototype but limits generalisation. For qualitative interviews, purposive sampling can deliberately select participants with relevant experience; adequacy is judged by information richness and saturation rather than a universal numerical rule.

A sample-size calculation should be linked to the intended analysis, effect size, variability, significance level, and desired power. For estimating a mean with a two-sided confidence interval of approximate half-width E, a common planning expression is:

Here n is the required number of independent observations, σ is the anticipated standard deviation, E is the desired margin of error, α is the significance level, and z_(1−α/2) is the corresponding standard-normal quantile. This is an approximation for a mean under suitable conditions. If σ is unknown, pilot data or a justified prior estimate may be used; finite-population, non-normal, clustered, or repeated-measures designs require appropriate adjustments.

In many engineering projects, the number of physical specimens is small. The researcher should then avoid claiming population-wide generalisation and instead report the tested units, operating envelope, uncertainty, and limitations.

5. Measurement, Uncertainty, and Data Quality

Measurement is a comparison between an observed quantity and a reference. Instruments have resolution, accuracy, repeatability, drift, bandwidth, loading, and calibration status. A displayed number with many digits is not necessarily precise.

Random error varies unpredictably between observations; systematic error produces a consistent bias. Repeating measurements can reduce uncertainty associated with random variation but cannot remove an unrecognised calibration bias. Before testing, inspect cables, connectors, grounding, probe attenuation, sensor polarity, sampling synchronisation, and instrument range.

For independent measured quantities x and y and a derived quantity q = f(x, y), a first-order uncertainty approximation is:

where u_x and u_y are standard uncertainties and covariance is neglected. If x and y are correlated, covariance terms must be included. The approximation is most suitable when uncertainties are small and the function behaves approximately linearly over the uncertainty range.

For example, if power P = VI and voltage and current uncertainties are independent, then:

Voltage V and current I are measured in volts and amperes, so P is in watts. This expression describes propagated relative standard uncertainty, not the full accuracy specification of an instrument system.

Data quality procedures should define missing-data handling, outlier rules, time synchronisation, filtering, and file naming before analysis. An outlier should not be deleted merely because it weakens a desired conclusion. Investigate whether it represents a genuine transient, instrument fault, protocol violation, or data-entry error, and document the decision.

6. Simulation, Verification, and Validation

Verification asks, “Was the model or implementation built correctly?” Validation asks, “Does it represent the real system adequately for the intended purpose?” A simulation can be verified against conservation laws, limiting cases, hand calculations, unit checks, and independent software. It can be validated by comparison with measured data under defined conditions.

A reproducible simulation record should include:

• governing equations or block-level model description;
• component and material parameters with sources;
• initial and boundary conditions;
• solver, time step, mesh, convergence tolerance, and stopping rule;
• random seeds where stochastic algorithms are used;
• software versions and operating environment;
• input files, scripts, and post-processing code;
• validation data and error metrics.

A useful prediction error metric is root mean square error:

Here N is the number of observations, yᵢ is the measured or reference value, and ŷᵢ is the predicted value. RMSE has the same units as y and penalises larger errors more strongly. It should not be used alone when scale, outliers, or asymmetric consequences matter. Mean absolute error, maximum error, bias, and confidence intervals may also be relevant.

Sensitivity analysis varies an input over a justified range and observes output changes. It identifies parameters that require accurate measurement and conditions under which conclusions may change. It is especially important when a model contains uncertain battery parameters, thermal coefficients, semiconductor losses, or user-behaviour assumptions.

7. Prototyping and Engineering Verification

Prototype development should follow requirements traceability. Each requirement receives an identifier, rationale, verification method, acceptance criterion, and result. Verification methods include inspection, analysis, simulation, test, and demonstration.

A typical workflow is:

1. define user, functional, performance, interface, environmental, and safety requirements;
2. develop alternative architectures and select one using explicit criteria;
3. calculate ratings, tolerances, thermal limits, protection needs, and worst-case conditions;
4. simulate critical circuits and control logic;
5. build and inspect the prototype;
6. perform low-energy and fault-safe tests before full-power operation;
7. compare results with acceptance criteria;
8. document deviations, revisions, and residual risks.

Design margins should be justified rather than assumed. Components must be assessed for voltage, current, power dissipation, temperature, insulation, creepage, clearance, switching stress, electromagnetic compatibility, and fault behaviour. A student prototype should never be connected to hazardous mains or high-energy storage without authorised supervision, appropriate isolation, protective equipment, emergency procedures, and institutional approval.

8. Data Analysis and Interpretation

Descriptive analysis should precede inferential analysis. Report sample size, mean, median, spread, range, plots, and raw or suitably anonymised observations. The sample mean is:

and the sample standard deviation is:

The n−1 denominator provides the usual unbiased estimate of population variance under independent random sampling from a population with finite variance. It is not a correction that makes biased or dependent measurements valid.

Statistical significance is not the same as engineering importance. A small difference may be statistically detectable but irrelevant to system requirements. Conversely, a practically important difference may be uncertain when the sample is small. Report effect sizes, uncertainty intervals, and requirement-based interpretation.

When comparing two approximately independent means with similar variances, a two-sample t procedure may be considered, subject to assumptions. If observations are paired—for example, the same motor tested before and after a controller update—a paired analysis is more appropriate. If distributions or variance assumptions are seriously unsuitable, transformations, robust methods, permutation procedures, or non-parametric methods may be considered. The selected method and assumptions must be justified rather than chosen because it produces a preferred p-value.

For classification systems, a confusion matrix distinguishes true positives, false positives, true negatives, and false negatives. Accuracy is:

where TP, TN, FP, and FN are the corresponding counts. Accuracy can be misleading for imbalanced fault datasets; precision, recall, specificity, F1 score, calibration, and cost-weighted performance may be more appropriate. Test data must be separated from training data to prevent leakage.

9. Reproducibility and Project Governance

Reproducibility means that another researcher can obtain consistent results by following the documented procedure with the stated materials and conditions. Repeatability concerns consistency under the same conditions; reproducibility often concerns consistency across operators, equipment, or locations. Definitions vary across disciplines, so the project should state how the terms are being used.

A reproducibility package may contain a protocol, bill of materials, circuit diagrams, firmware, source code, simulation files, raw data, metadata, analysis scripts, calibration records, and a README. Use version control, meaningful commit messages, tagged releases, and automated or scripted analysis where feasible. Record units, time zones, sensor serial numbers, firmware versions, and changes to the procedure.

The proposal should include a schedule with milestones, dependencies, risk owners, and contingency actions. A risk register can rank likelihood and consequence, but numerical rankings do not replace engineering judgement. Typical risks include component delays, unstable prototypes, unavailable instruments, corrupted data, insufficient participants, and scope growth.

10. Ethics, Safety, and Responsible Engineering

Electrical-engineering research has ethical responsibilities even when it does not involve human subjects. Safety risks include electric shock, arc flash, fire, rotating machinery, stored energy, laser or radio-frequency exposure, toxic materials, and battery failure. Risk assessment, supervision, lockout or isolation procedures, protective equipment, emergency response, and inspection are part of the methodology.

Human-participant work requires informed consent, voluntary participation, privacy protection, minimisation of harm, secure data storage, and appropriate institutional review. A user study should not collect identifiable data unnecessarily. Participants must understand the purpose, procedures, risks, benefits, withdrawal rights, and use of recordings or quotations.

Research integrity requires accurate reporting, proper attribution, preservation of raw data, disclosure of conflicts of interest, and clear distinction between measured, simulated, assumed, and inferred results. Fabrication, falsification, plagiarism, selective exclusion, undisclosed image manipulation, and inappropriate reuse of data are unacceptable.

Artificial intelligence can assist with literature discovery, coding, data inspection, or language editing, but generated content must be verified. A model may invent references, introduce unsafe code, conceal assumptions, or reproduce confidential material. Students remain responsible for technical decisions, citations, data protection, and disclosure of significant AI assistance according to institutional policy.

11. Preparing the Project Proposal

A concise proposal should contain the following connected elements:

1. Title and abstract: specific enough to identify system, method, and outcome.
2. Background and problem: explains context, importance, and consequences.
3. Literature review: synthesises established knowledge, methods, limitations, and the gap; it should not be an annotated list.
4. Aim, objectives, and questions: aligned and measurable.
5. Conceptual framework: identifies variables, mechanisms, assumptions, and boundaries.
6. Methodology: gives design, apparatus, participants or specimens, sampling, controls, procedure, and analysis.
7. Metrics and acceptance criteria: defines how success will be judged before results are known.
8. Verification and validation plan: specifies reference cases, calibration, comparison, and uncertainty treatment.
9. Safety, ethics, and data management: identifies approvals, hazards, access controls, retention, and backup.
10. Resources and schedule: includes equipment, software, budget, responsibilities, milestones, and contingencies.
11. Expected contribution and limitations: states what the project will add and what it will not establish.
12. References: uses a consistent scholarly style and includes primary or authoritative sources.

Alignment is the central quality test. Each objective should have a method, each method should generate data, each dataset should support an analysis, and each analysis should answer a question. A proposal that promises a “real-time, low-cost, highly accurate, universally applicable” system without defining real-time, cost, accuracy, or operating scope is not yet research-ready.

Applied Case: Illustrative Battery State-of-Charge Estimator

Background

An undergraduate team proposes a state-of-charge estimator for a lithium-ion battery used in a small electric vehicle. The practical problem is that coulomb counting drifts because of sensor bias and uncertain initial charge, while open-circuit-voltage methods respond slowly and require rest periods.

Problem and objectives

The team asks whether a Kalman-filter-based estimator can maintain an absolute state-of-charge error below 5% over a defined current profile, when compared with a laboratory reference procedure. Objectives are to identify a battery model, implement the estimator on a microcontroller, measure current and voltage, and evaluate error, execution time, and robustness to sensor bias.

State of charge is represented as a fraction between 0 and 1. A simplified coulomb-counting update is:

Here SOCₖ is the estimated state of charge at sample k, η_c is coulombic efficiency, Δt is the sampling interval in seconds, Qₙ is nominal capacity in ampere-seconds, and Iₖ is signed current in amperes, defined positive for discharge. The model assumes the sign convention is applied consistently and that capacity and efficiency are suitably known over the tested conditions. It does not by itself account for temperature dependence, rate dependence, ageing, or voltage relaxation.

Design

The independent conditions are current-profile type and imposed sensor-bias level. Dependent variables are absolute SOC error, maximum error, computation time, and memory use. The baseline is coulomb counting with a measured initial SOC. The proposed estimator is tested using the same battery, profile, sampling interval, and instrumentation. Separate charge and discharge profiles are used, and the reference SOC is obtained using a documented laboratory method with uncertainty stated.

The team uses repeated runs on multiple battery units where possible, randomises profile order to reduce ageing-order effects, logs raw voltage, current, temperature, timestamps, firmware version, and filter parameters, and reserves one profile for final evaluation. The estimator is not tuned on the final test profile.

Analysis and outcome criteria

The primary metric is mean absolute error:

SOCᵢ is the reference value and SOĈᵢ is the estimate at observation i. MAE is dimensionless and may be reported as a percentage by multiplying by 100. The acceptance criterion is MAE ≤ 0.05 over the declared test envelope, together with a maximum error limit and a stated computation-time limit. The team reports confidence intervals or run-to-run variation and does not claim performance outside the tested temperature, current, battery chemistry, age, and capacity range.

Learning points

The case illustrates that a method is defensible because the question, baseline, variables, reference measurement, test profile, metrics, and limits are aligned. It also demonstrates why a formula can be correct yet insufficient: the coulomb-counting equation requires assumptions about capacity, efficiency, bias, and initial state. The project’s contribution is an evaluated estimator within a declared envelope, not proof that one algorithm is universally best.

Challenges and Limitations

Engineering projects face trade-offs between scope, realism, cost, and control. Highly controlled tests may have strong internal validity but limited similarity to field conditions. Field data may be realistic but contain uncontrolled confounding. Simulations enable broad exploration but inherit model assumptions. Prototypes reveal integration problems but may have limited repeatability.

Small student projects often have too few specimens to support strong statistical generalisation. Instrument availability may constrain sampling frequency or uncertainty. Components may vary between batches. Software updates can change results. Time pressure can encourage post hoc method changes. These limitations should be stated and managed through pilot testing, preregistered or time-stamped analysis plans where appropriate, sensitivity analysis, triangulation, and careful claims.

A useful conclusion distinguishes four levels: what was directly measured, what was inferred under assumptions, what was demonstrated only in simulation or prototype form, and what remains unknown. This distinction is a mark of engineering maturity rather than weakness.

Best-Practice Checklist

Before submitting a proposal, confirm that:

• the problem and contribution are bounded;
• objectives and research questions are measurable;
• the baseline or control is credible;
• variables and operational definitions are explicit;
• assumptions and limitations are stated;
• sampling and replication are justified;
• instruments and calibration are identified;
• procedures can be followed by another person;
• metrics and acceptance criteria are defined before testing;
• simulation is verified and, where possible, validated;
• safety, ethics, privacy, and intellectual property are addressed;
• raw data, code, versions, and decisions will be preserved;
• conclusions will not exceed the evidence.

Chapter Summary

Research design links an engineering question to evidence capable of supporting a conclusion. A defensible project begins with a bounded problem, aligned aim, measurable objectives, and explicit research questions. Quantitative, qualitative, experimental, simulation, prototyping, and mixed methods each answer different kinds of questions and may be combined when their roles are clear.

Variables, assumptions, controls, sampling, operational definitions, metrics, and acceptance criteria determine the quality of evidence. Experiments require suitable controls and replication; simulations require verification and validation; prototypes require requirements traceability and safety management. Measurement quality depends on calibration, uncertainty, bandwidth, synchronisation, and transparent treatment of anomalies. Statistical significance must be distinguished from engineering importance.

Reproducibility requires documented procedures, versioned code, raw data, metadata, and parameter provenance. Ethical engineering includes safe practice, responsible human-participant research, honest reporting, data protection, and critical verification of AI-assisted work. A strong proposal aligns every objective with a method, dataset, analysis, and answer while clearly stating scope and limitations.

Key Terms

Research design: The structured plan for obtaining and interpreting evidence.

Engineering methodology: A systematic process for solving and evaluating an engineering problem.

Paradigm: A view of knowledge and acceptable methods of inquiry.

Independent variable: A factor deliberately changed or used as a predictor.

Dependent variable: The measured outcome of interest.

Control variable: A condition held constant or balanced to reduce confounding.

Confounder: A variable that distorts an apparent relationship between variables.

Operational definition: A precise description of how a construct is measured.

Hypothesis: A testable prediction about a relationship or difference.

Baseline: A reference method, system, or condition used for comparison.

Replication: Repeating a procedure under defined conditions.

Randomisation: Assigning conditions or order by chance to reduce systematic bias.

Validity: The degree to which a method supports the intended claim.

Reliability: The consistency of a measurement or procedure.

Verification: Checking whether an implementation or model was built correctly.

Validation: Assessing whether a model or system represents the intended real-world behaviour.

Uncertainty: Quantified doubt associated with a measurement or result.

Reproducibility: The ability to obtain consistent results using documented materials and procedures.

Requirements traceability: Linking each requirement to its design, verification method, and result.

Mixed methods: Planned integration of quantitative and qualitative evidence.

Review Questions

Short Answer Questions

1. Distinguish an engineering aim from a research objective.
2. What is the difference between verification and validation?
3. Why is a baseline important in an experimental engineering study?
4. Define an operational definition and give one electrical-engineering example.
5. State two limitations of convenience sampling.

Long Answer Questions

1. Compare experimental, simulation, and prototyping methodologies for evaluating a motor-drive controller. Discuss evidence, controls, validation, and limitations.
2. Explain how variables, assumptions, controls, and confounders should be identified in a power-electronics experiment.
3. Describe a reproducible workflow for a simulation-based electrical-engineering project.
4. Explain how uncertainty, calibration, repeatability, and measurement bandwidth affect engineering conclusions.
5. Develop the recommended structure of a major-project proposal and explain how alignment should be demonstrated.

Analytical / Application Questions

1. Design a test matrix to compare two inverter-control algorithms across three load levels and two ambient temperatures. Specify controls, replication, metrics, and a method for handling run order.
2. A fault-classification system reports 98% accuracy on a dataset in which only 2% of cases are faults. Analyse why accuracy may be misleading and propose more suitable evaluation measures and a leakage-control procedure.
3. Prepare a methodology for evaluating the battery state-of-charge estimator in the illustrative case. Identify the reference method, variables, assumptions, acceptance criteria, safety controls, and limitations.

Multiple Choice Questions with Answers

Question 1

Which statement best describes research design in an engineering project?

A. A list of components required to build a prototype

B. A structured plan linking a research question to evidence and analysis

C. A final presentation of project results

D. A software package used for statistical testing

Correct Answer: B. A structured plan linking a research question to evidence and analysis

Question 2

In a converter study, output efficiency is measured while switching frequency is deliberately changed. Switching frequency is the:

A. Dependent variable

B. Control variable

C. Independent variable

D. Confounding measurement

Correct Answer: C. Independent variable

Question 3

What is the primary purpose of a baseline?

A. To increase the number of components in a prototype

B. To provide a reference for judging the proposed method

C. To eliminate all measurement uncertainty

D. To replace calibration procedures

Correct Answer: B. To provide a reference for judging the proposed method

Question 4

Which activity is validation?

A. Checking a simulation against a hand calculation for a limiting case

B. Confirming that source code compiles

C. Comparing model predictions with measured system behaviour

D. Inspecting whether a circuit diagram has labels

Correct Answer: C. Comparing model predictions with measured system behaviour

Question 5

If P_out = 90 W and P_in = 100 W, the converter efficiency is:

A. 9%

B. 10%

C. 90%

D. 190%

Correct Answer: C. 90%

Question 6

Which metric is often more informative than accuracy for a highly imbalanced fault-detection dataset?

A. Precision and recall

B. File size

C. Nominal supply voltage

D. Number of software comments

Correct Answer: A. Precision and recall

Question 7

What does replication primarily help assess?

A. Whether random variation and repeatability affect results

B. Whether a research question is grammatically correct

C. Whether a component has a manufacturer logo

D. Whether a model has no assumptions

Correct Answer: A. Whether random variation and repeatability affect results

Question 8

Which practice most directly supports reproducibility?

A. Reporting only the best experimental run

B. Changing parameters without recording them

C. Preserving raw data, code, versions, and procedure details

D. Using rounded values without units

Correct Answer: C. Preserving raw data, code, versions, and procedure details

Question 9

In the equation η = (P_out / P_in) × 100%, η represents:

A. Temperature rise

B. Efficiency

C. Sampling frequency

D. Measurement bias

Correct Answer: B. Efficiency

Question 10

Why must engineering conclusions state their operating envelope?

A. A prototype result automatically applies to every possible system

B. Performance may depend on tested conditions such as temperature, load, and component variation

C. Operating conditions are irrelevant after calibration

D. It prevents any comparison with a baseline

Correct Answer: B. Performance may depend on tested conditions such as temperature, load, and component variation