Gains:
- Ability to holistically evaluate the ethical, data privacy, copyright and workforce aspects of AI use in textile
- Ability to design a system with designated human approval points that combines the entire module in an end-to-end workflow
- Ability to establish a governance, validation and continuous improvement framework for responsible AI use in an organization
Module Exam
1. Which of the following is the basic principle that determines the risk of an AI output in textile engineering?
- A) The risk of an output is equal to the harm it would cause if that output were faulty; validation scales accordingly ✔
- B) AI output is generally safe as it is written fluently and confidently
- C) Validation is unnecessary when the most current model is used
- D) If the output refers to a standard, no additional control is required
Explanation: The risk of an output is equal to the harm it will cause if that output is faulty. While a mistake in a moodboard idea is harmless, a mistake in a dyeing recipe or fabric strength calculation can lead to batch waste and occupational safety/contract risks; so verification intensity scales with potential harm.
2. An engineer is skeptical of the AI value of '85 cN/tex for cotton yarn' for yarn strength. What is the quickest check to confirm this suspicion?
- A) Put the value directly on the report and ask the laboratory later
- B) Make an order of magnitude comparison with the typical cotton yarn range (approximately 12-22 cN/tex) ✔
- C) Asking the same question from the model again and trusting the second answer
- D) Accept if the value contains a reference to a source
Description: It is an order of magnitude control. Strength in ring cotton yarn is typically on the order of 12-22 cN/tex; 85 cN/tex is the high performance synthetic/aramid level and is physically unreasonable for cotton. This check takes seconds and catches most unit/type errors.
3. Which of the following is true for 'Ne' (British cotton count) among the yarn count systems?
- A) The larger the thread becomes, the thicker it becomes.
- B) What is a direct system, it works in the same direction as tex
- C) What an indirect system it is; The larger the number, the thinner the thread ✔
- D) What is defined only for synthetic yarns
Explanation: What an indirect numbering system: the larger the number, the thinner the thread. Tex and denier are direct systems; The larger the number, the thicker the thread. Not knowing this distinction leads to serious errors in number interpretations in the AI output.
4. What does it mean to have a high 'false negative' rate in an image processing model that detects fabric defects and why is it critical?
- A) The defective fabric is deemed perfect and delivered to the customer; ✔ The most dangerous type of error in quality
- B) Unnecessary marking of perfect fabric as defective; it just reduces efficiency
- C) The model finds no faults; always a good sign
- D) Low resolution of the camera
Explanation: A false negative is when a true defect is classified as 'perfect'. This means that the faulty fabric goes to the customer and incurs return/reputation costs. In quality-critical applications, the threshold is generally set to keep false negatives low, and increased false positives are eliminated by human inspection.
5. What does the 'Delta E' (ΔE) value mean in color management?
- A) Cost per kilogram of paint
- B) The criterion that quantifies the total perceptual difference between two colors ✔
- C) Light transmittance of the fabric
- D) Value of the dyeing temperature in degrees
Description: Delta E is the criterion that quantifies the total perceptual difference between two colors in the CIELAB color space. Small ΔE values indicate that the colors are close together; In textiles, batch-to-batch acceptance tolerance is usually defined contractually in terms of ΔE. Therefore, the claim that 'the color held' requires measurement.
6. What is it called when two samples look the same under one light source and different under another light source, and why is it important in color management?
- A) Metamerism; Color confirmation must be done under multiple standard light sources ✔
- B) Fluorescence; color looks the same in all conditions
- C) Opacity; color difference only depends on thickness
- D) Calibration; only concerns the screen
Description: This phenomenon is called metamerism. Two colors with different spectral reflectance may match under certain illumination and diverge in other illumination. Therefore, color confirmation should not be done on a single light source, but under multiple sources (e.g. D65, TL84, A) in a standard light booth.
7. What is the most critical legal check to be made before commercial use of a textile pattern produced with generative AI?
- A) Reducing the number of colors of the pattern
- B) Reducing the file size of the pattern
- C) Checking for similarity/infringement and copyrights with existing protected designs ✔
- D) Save the pattern only in high resolution
Explanation: It is necessary to check whether the produced pattern bears any similarity/infringement with existing protected designs, trademarks or copyrighted works, and to evaluate the uncertainties regarding the model's training data and output rights. Commercialization without checking originality and infringement creates the risk of copyright litigation.
8. What is the most robust approach to validating AI output in forecasting demand by SKU?
- A) Converting the single point prediction given by the model into the order quantity as it is
- B) Backtesting with past sales and evaluating the margin of error through business judgment filter ✔
- C) Choosing the highest estimate and keeping excess stock
- D) Double the prediction if the trend is the most popular
Explanation: The forecast should be backtested against past sales realization and filtered through business judgment with context such as inventory, campaigns, seasonality, etc. Rather than blindly trusting a single point prediction of the model, the margin of error and scenarios are evaluated.
9. What is the main reason for the paint batch order from 'light color to dark color' in production scheduling?
- A) Dark colors are cheaper
- B) Reducing the risk of color contamination and machine cleaning times ✔
- C) Light colors last longer
- D) It is only an aesthetic preference
Explanation: If a light color is dyed after a dark color in the same machine, the risk of cleaning/contamination and additional washing time increases. Sorting from light to dark increases efficiency by reducing the risk of color smearing and machine cleaning times. AI recommendations should include this process constraint.
10. When reporting the environmental footprint of a fabric, what is the correct attitude towards a claim such as 'recycled polyester reduces water use by 90%' produced by AI?
- A) Putting the claim as it is in the marketing copy
- B) Not using the claim without verification with system boundary, data source and certificate ✔
- C) Assuming the claim is strong if the number is high
- D) Round the number and publish it
Explanation: Such a claim is meaningless without specifying the system boundary (cradle-to-gate or gate-to-gate), data source and scope; It must be based on verifiable LCA data and certification. Otherwise, greenwashing and legal risk may arise. While AI numbers seem reasonable, they are independently sourced.
11. What should be the first step to verify the code written by the AI when analyzing production data using pandas with Python?
- A) If the code worked without errors, accept the result as correct
- B) Manually calculate the output and verify the unit and logic on a small data sample ✔
- C) Adding the code directly to the production report
- D) If the graph turns out good, consider the analysis completed.
Remark: Even if the code works, it is not guaranteed that the logic is correct; Manual accounting check must be done on unit, column matching and several rows. For example, when averaging grammage, incorrect grouping or unit confusion may silently produce incorrect results. In a small sample, manual verification is the fastest assurance.
12. What is the best practice in terms of data privacy when sending textile production data to an AI tool?
- A) Send all actual customer and prescription information as is
- B) Anonymize sensitive information and send it with representative labels and in accordance with corporate policy ✔
- C) Assuming that privacy is unimportant because there is so much data.
- D) Just change the file name and send the content as it is
Explanation: Sensitive data such as customer name, recipe secret, price and supplier information should be anonymized or given with representative labels (Customer-A, Recipe-1). Trade secrets and personal data should not be transferred to a third-party model in violation of corporate policies and agreements.
13. An image processing model reports 98% accuracy on a production line but misses many defects in the field. What is the most likely cause of this?
- A) Accuracy is misleading due to unbalanced data; The sensitivity (recall) of the defect class is low ✔
- B) The model is running very slowly
- C) Because the camera is color
- D) 98% is already perfect, the problem is with the operator
Explanation: When samples with defects are very few in the data set (imbalanced data), the model can achieve high accuracy by calling everything 'perfect' but cannot capture the actual defects. Therefore accuracy alone is misleading; Sensitivity (recall) and precision for the defect class should be monitored.
14. According to the key principle emphasized throughout the module, what is the role of AI in safety and contract-critical textile decisions?
- A) AI can make critical decisions on its own because it is fast
- B) AI is an assistant; critical decisions are verified by competent engineer approval and physical testing ✔
- C) AI output replaces the signed report
- D) Human confirmation is unnecessary when the most current model is used
Description: AI is an assistant and accelerator; generates hypotheses, prepares drafts, catches patterns. However, critical decisions such as painting recipe approval, quality acceptance, strength/safety evaluation require competent engineer approval and physical testing. AI output is not a substitute for expert approval.