QCM : Automated PCB Fault Diagnosis — 22 questions

Questions et réponses du QCM

1. What best describes a printed circuit board (PCB) in terms of its physical structure and function?

A metal frame that holds components and performs electrical testing
A software system that uses computer vision to inspect PCBs for faults
A rugged glass-reinforced plastic board with etched copper tracks that connect components into circuits
A camera and lighting setup that captures PCB images for diagnosis

A rugged glass-reinforced plastic board with etched copper tracks that connect components into circuits

Explication

A PCB is a physical glass-reinforced plastic board with etched copper tracks that interconnect components into circuits. The other options describe inspection tools or systems rather than the PCB itself.

2. In PCB connectivity failures, which pair correctly distinguishes open circuits from short circuits?

Open circuits only affect component polarity, while short circuits only affect solder bridges
Open circuits break a connection path, while short circuits create an unintended connection
Open circuits and short circuits are the same defect described with different names
Open circuits create an unintended connection, while short circuits break a connection path

Open circuits break a connection path, while short circuits create an unintended connection

Explication

An open circuit breaks the intended connection path, whereas a short circuit creates an unintended electrical connection. The distractors swap the definitions or incorrectly claim they are interchangeable labels.

3. What is the main purpose of using Roboflow in the research workflow?

To compute economic maintenance-cost estimates for repair decisions
To perform thermal imaging and X-ray inspection of PCBs
To support image annotation, dataset management, preprocessing, and model training for a vision fault-diagnosis system
To replace manual inspection equipment in manufacturing

To support image annotation, dataset management, preprocessing, and model training for a vision fault-diagnosis system

Explication

The study uses Roboflow to carry out annotation, dataset management, preprocessing, and model training. The other options refer to excluded evaluation methods or unrelated system roles.

4. If a model improves mean Average Precision (mAP), which interpretation matches the study’s evaluation framing?

mAP is an object-detection metric, while accuracy, precision, recall, and F1-score summarize classification performance
accuracy, precision, recall, and F1-score are detection-only metrics, while mAP is for classification
mAP is a classification metric identical to F1-score
mAP is used only for inference-time measurement, not for performance quality

mAP is an object-detection metric, while accuracy, precision, recall, and F1-score summarize classification performance

Explication

The study treats mAP as an object-detection metric, while accuracy/precision/recall/F1-score summarize classification performance. The distractors incorrectly swap metric meanings or misassign them to inference time.

5. How does the study’s evaluation scope position automated inspection relative to conventional manual inspection?

As an economic-cost evaluation of repair budgets
As a three-dimensional and thermal analysis of PCB materials
As a comparative assessment focused on defect-detection accuracy, inspection time, and consistency
As a replacement of manufacturing hardware rather than an evaluation

As a comparative assessment focused on defect-detection accuracy, inspection time, and consistency

Explication

The study compares automated computer-vision inspection with conventional manual inspection using accuracy, inspection time, and consistency. The other options describe scope exclusions or different evaluation types.

6. In the Roboflow workflow, what does augmentation primarily do compared with annotation?

It automatically creates bounding boxes without requiring any labels
It changes training examples but does not replace the need to label defect locations and classes
It exports the dataset without affecting how images are used during training
It merges training and testing sets to increase dataset size

It changes training examples but does not replace the need to label defect locations and classes

Explication

Annotation identifies defect locations and classes, while augmentation modifies training examples without replacing labels. A distractor that claims augmentation can substitute for labeling is contrary to the workflow description.

7. To keep an evaluation unbiased in a PCB defect study, what should you do with an image that you plan to use for testing if it was also used during training?

Exclude it from the training set so it is only used for testing
Use it only for testing, but you may reuse the same image for data augmentation in training
Keep it in training and also include it in testing because the model sees it during optimization
Reuse it in training and testing as long as the model was not retrained

Exclude it from the training set so it is only used for testing

Explication

For unbiased evaluation, a testing image must not also be used for training. The model should be evaluated on images that were not part of training, so option 1 is correct while the others violate this separation.

8. In visual inspection, what best describes how a convolutional neural network (CNN) is used?

A rule-based system that detects defects using fixed thresholds and no feature learning
A method that uses only manually designed features without learning image representations
A deep-learning architecture that learns complex visual defect features automatically from images
A traditional computer-vision pipeline that relies primarily on hand-crafted defect templates

A deep-learning architecture that learns complex visual defect features automatically from images

Explication

A CNN is a deep-learning architecture designed to learn complex visual defect features automatically from images. Traditional image processing typically depends more on manually designed features.

9. What is the key reason YOLO, Faster R-CNN, and RetinaNet are described as having different trade-offs?

They are identical architectures that mainly vary in naming and evaluation datasets
They differ only in training time and not in inference speed or accuracy
They offer different balances between object-detection inference speed and accuracy
They guarantee the same accuracy for all datasets while varying only runtime

They offer different balances between object-detection inference speed and accuracy

Explication

YOLO, Faster R-CNN, and RetinaNet are characterized by different trade-offs between inference speed and detection accuracy. A faster model is not automatically the more accurate one.

10. Which set of issues best matches the kinds of research gaps identified in the literature?

Full reporting of all metrics and complete multi-class coverage across diverse datasets
High-quality dataset generalization, broad deployment evaluation coverage, and consistently simple toolchains
Limited Roboflow integration, limited deployment evaluation, dependence on complex frameworks, and incomplete metric reporting
No dependence on frameworks because only Roboflow is used for end-to-end deployment

Limited Roboflow integration, limited deployment evaluation, dependence on complex frameworks, and incomplete metric reporting

Explication

The identified research gaps include items such as limited Roboflow integration, insufficient deployment evaluation, complex frameworks, limited multi-class capability, and incomplete metric reporting. The distractors contradict these gap themes.

11. In an object-detection study pipeline for PCB defects, what sequence best matches the described methodology?

Use only a pre-trained detector with no annotation step and evaluate using qualitative observation
Train first, then collect images afterward, and apply preprocessing only after dataset splitting
Annotate only during testing, train directly on the validation set, and skip quantitative evaluation
Collect images, preprocess and annotate in Roboflow, split into training/validation/testing, train a detector, and evaluate quantitatively

Collect images, preprocess and annotate in Roboflow, split into training/validation/testing, train a detector, and evaluate quantitatively

Explication

The methodology collects defective and non-defective images, preprocesses/annotates them in Roboflow, splits the data, trains the detector, and evaluates it quantitatively. The alternatives either swap steps or omit required steps like annotation and quantitative evaluation.

12. Which description best fits the recommended experimental software environment for the study?

Python 3.10+ with Roboflow, OpenCV, NumPy, Pandas, Ultralytics YOLO, Matplotlib, Scikit-learn, PyTorch, and Jupyter or VS Code
Matlab with a built-in YOLO detector, plus manual spreadsheets for data handling
A single-tool setup limited to Roboflow and Matplotlib without any deep-learning framework
Only Python 3.10+ and a camera, with no need for Roboflow, OpenCV, or machine-learning libraries

Python 3.10+ with Roboflow, OpenCV, NumPy, Pandas, Ultralytics YOLO, Matplotlib, Scikit-learn, PyTorch, and Jupyter or VS Code

Explication

The recommended environment includes Python 3.10 or later plus Roboflow, OpenCV, NumPy, Pandas, Ultralytics YOLO (YOLOv8 or YOLOv11), Matplotlib, Scikit-learn, PyTorch, and Jupyter Notebook or Visual Studio Code.

13. How is accuracy calculated in terms of TP, TN, FP, and FN?

2 × (Precision × Recall) / (Precision + Recall)
(TP + TN) / (TP + TN + FP + FN)
TP / (TP + FN)
TP / (TP + FP)

(TP + TN) / (TP + TN + FP + FN)

Explication

Accuracy is defined as (TP + TN) / (TP + TN + FP + FN). The distractors correspond to precision, recall, and F1-score formulas instead.

14. If the goal is to measure how many predicted defects are actually defects, which metric is being captured?

Accuracy
Precision
Recall
F1-score

Precision

Explication

Precision is calculated as TP / (TP + FP) and reflects the proportion of predicted defects that are truly defects. Recall instead focuses on detecting real defects.

15. Which metric indicates how many real defects the system detects?

Specificity
Precision
Accuracy
Recall

Recall

Explication

Recall is computed as TP / (TP + FN) and represents the fraction of real defects that are detected. Precision measures correctness of predicted defects rather than coverage of real defects.

16. What is the F1-score designed to do when precision and recall trade off?

Balance precision and recall using the harmonic mean to help with class imbalance
Maximize recall only by reducing false negatives at all costs
Maximize precision only by reducing false positives at all costs
Replace accuracy so that true negatives are no longer considered

Balance precision and recall using the harmonic mean to help with class imbalance

Explication

The F1-score is 2 × (Precision × Recall) / (Precision + Recall), balancing precision and recall, which is especially helpful when defect classes are imbalanced. The distractors either optimize only one side or misstate what the score is used for.

17. What does the inspection flow do to each PCB image before running detection?

It captures a PCB image, resizes, enhances, normalizes, and crops it, then sends it through the Roboflow pipeline and the trained model for detection and fault classification.
It captures the PCB image, performs detection first, and then normalizes and crops the image afterward.
It only captures the PCB image and directly runs the trained model without changing size or image quality.
It captures the PCB image, resizes it, and stores it without enhancement, then skips the Roboflow pipeline.

It captures a PCB image, resizes, enhances, normalizes, and crops it, then sends it through the Roboflow pipeline and the trained model for detection and fault classification.

Explication

The workflow captures the PCB image, preprocesses it (resizing/enhancing/normalizing/cropping), then sends it through the Roboflow pipeline for detection using the trained model. The other options either skip key preprocessing steps or omit the pipeline/model detection flow.

18. When the system finds no defect, what output should it display?

Bounding boxes, class labels, and confidence scores
Only the class label with no confidence score
A stored image preview without any detection results
No Defect

No Defect

Explication

The system displays “No Defect” when no defect is detected. The bounding-box/class/score output corresponds to cases where defects are detected.

19. In the proposed inspection workflow, what role does Roboflow primarily play?

It replaces the trained detection model by performing fault classification only through rules.
It guarantees perfect detection accuracy regardless of dataset quality and image conditions.
It physically measures PCB dimensions and calibrates the camera for accurate detection.
It simplifies image annotation, dataset management, augmentation, model training, and deployment.

It simplifies image annotation, dataset management, augmentation, model training, and deployment.

Explication

Roboflow is used to streamline annotation, dataset management, augmentation, training, and deployment in the workflow. The distractors describe functions not stated as Roboflow’s purpose in the workflow.

20. Which statement best reflects the system’s reported advantage over traditional human visual inspection?

It relied on manual labeling during inference instead of using trained models.
It detected and classified multiple PCB defects more rapidly and consistently.
It required inspectors to review every image only after long offline processing.
It achieved fewer classifications by limiting itself to single-defect detection.

It detected and classified multiple PCB defects more rapidly and consistently.

Explication

The system processed images more rapidly and consistently while detecting and classifying multiple defects compared with human visual inspection. The other options contradict the described speed/consistency improvement.

21. Which factor most directly affects detection performance in the developed system?

The color of the detector overlay used for bounding boxes.
The quality and diversity of annotated training data, including image quality and lighting.
The presence of a fixed-size cropping window regardless of input image quality.
The physical thickness of the PCB board materials.

The quality and diversity of annotated training data, including image quality and lighting.

Explication

Detection performance depends on high-quality annotated data and factors like dataset diversity and image/lighting conditions. The other options are not listed as determinants of performance.

22. What set of future improvements is proposed to advance the system?

Expand datasets, compare detection architectures, enhance preprocessing and augmentation, integrate industrial equipment, retrain continuously, and evaluate multi-class and severity-based classification.
Limit the system to a single detection architecture and remove augmentation to keep results stable.
Replace classification with manual inspection after every detection while keeping the model fixed.
Only reduce preprocessing steps, stop retraining once models are trained, and avoid changes to dataset size.

Expand datasets, compare detection architectures, enhance preprocessing and augmentation, integrate industrial equipment, retrain continuously, and evaluate multi-class and severity-based classification.

Explication

The future development plan includes expanding datasets, comparing architectures, improving preprocessing/augmentation, integrating industrial inspection equipment, continuously retraining, and evaluating multi-class/severity-based classification. The distractors either remove key improvement steps or conflict with the proposed continuous retraining and evaluation directions.

Révisez avec les flashcards

Mémorisez les réponses avec 56 flashcards sur Automated PCB Fault Diagnosis.

What is a printed circuit board made of?

Glass-reinforced plastic with etched copper tracks.

Why are PCBs used in almost all commercial electronic devices?

Because they are rugged, inexpensive, and highly reliable.

What are the main steps in bare PCB fabrication?

Copper foil coverage, photoresist film formation, etching, and film stripping.

Voir les flashcards →

Approfondir avec la fiche

Consultez la fiche de révision complète sur Automated PCB Fault Diagnosis.

Voir la fiche →

Cours similaires

Gestion des fichiers en PHP

Programmation

Identification utilisateur en PHP

Programmation

Rôle d'un système d'exploitation

Questions techniques projet Compawgnon

Bases de Python pour débutants

Programmation

Cours complet de cybersécurité

Cybersécurité

Crée tes propres QCM

Importe ton cours et l'IA génère des QCM avec corrections en 30 secondes.

Générateur de QCM