SACA Smart Factory Data Analytics 5 — Questions and Answers
Question 1: Which smart factory data strategy involves collecting raw data without a predefined schema to enable flexible future analysis?
- Schema-on-write
- Schema-on-read (Correct answer)
- Structured data warehousing
- Normalized relational modeling
Correct answer: Schema-on-read
Schema-on-read allows raw data to be stored in a data lake without enforcing structure at ingestion, applying the schema only when the data is queried.
Question 2: A factory wants to predict remaining useful life (RUL) of a motor using sensor history. Which model family is best suited for learning temporal dependencies in the degradation signal?
- Decision tree
- Naive Bayes classifier
- Recurrent Neural Network (RNN) / LSTM (Correct answer)
- K-nearest neighbors
Correct answer: Recurrent Neural Network (RNN) / LSTM
RNNs and LSTMs are designed to capture long-range temporal dependencies in sequential data, making them well-suited for RUL prediction from sensor time-series.
Question 3: What does the term 'cold path' refer to in a lambda architecture deployed in a smart factory?
- Real-time processing of sensor streams with sub-second latency
- Batch processing of historical data for deep analytics and model retraining (Correct answer)
- Encrypted transmission of data to cold storage
- Shutdown procedure for equipment during planned downtime
Correct answer: Batch processing of historical data for deep analytics and model retraining
In lambda architecture, the cold (batch) path processes large volumes of historical data at high latency to produce accurate, comprehensive analytics.
Question 4: Which data governance practice ensures that factory sensor data can be traced from its origin machine through every transformation to its final analytics output?
- Data masking
- Data lineage tracking (Correct answer)
- Data sharding
- Data partitioning
Correct answer: Data lineage tracking
Data lineage tracking records the full journey of data from source to output, enabling auditability, debugging, and regulatory compliance.
Question 5: A quality control model achieves 99% accuracy on a dataset where 99% of parts are non-defective. Why is this accuracy misleading?
- The model is overfitting to the training set
- The high accuracy reflects class imbalance, not genuine defect detection ability (Correct answer)
- Accuracy is not a valid metric for classification tasks
- The dataset is too small for meaningful evaluation
Correct answer: The high accuracy reflects class imbalance, not genuine defect detection ability
When the negative class dominates, a model that always predicts 'non-defective' achieves high accuracy while completely failing to detect any real defects.
Question 6: In smart factory analytics, what is 'feature engineering' in the context of preparing sensor data for machine learning?
- Selecting which physical sensors to install on the factory floor
- Transforming raw sensor readings into meaningful input variables that improve model performance (Correct answer)
- Writing firmware for embedded sensor microcontrollers
- Configuring network routing for sensor data streams
Correct answer: Transforming raw sensor readings into meaningful input variables that improve model performance
Feature engineering transforms raw measurements into derived signals (e.g., rolling mean, FFT amplitude, rate of change) that encode domain knowledge and improve model accuracy.
Question 7: Which standard communication model allows different smart factory devices from multiple vendors to exchange data using a unified information model?
- Modbus RTU
- OPC UA (OPC Unified Architecture) (Correct answer)
- RS-232 serial
- CAN bus
Correct answer: OPC UA (OPC Unified Architecture)
OPC UA provides a platform-independent, vendor-neutral information model and secure communication framework adopted widely across industrial automation and smart factory systems.
Which smart factory data strategy involves collecting raw data without a predefined schema to enable flexible future analysis?