Digital Materials Engineering, Analytics and AI

Introduction and Context

Construction materials are increasingly managed as information-rich assets rather than as isolated physical products. A concrete mix, structural steel member, geopolymer panel or waterproofing membrane has a history: constituent sources, batch records, test results, transport conditions, installation data, exposure history and maintenance interventions. Digital materials engineering connects this information across the asset life cycle and supports decisions about selection, quality assurance, traceability, durability and replacement.

The principal technologies are building information modelling (BIM), material databases, internet-of-things sensors, digital twins, computer vision, machine learning and predictive analytics. Their value does not arise from digitisation alone. It depends on valid data, suitable models, transparent decision rules, competent users and governance of safety-critical outcomes. An inaccurate digital model can produce decisions that are more systematically wrong than decisions based on limited but reliable records.

This chapter examines how digital methods can be integrated into advanced construction-materials practice. It also addresses uncertainty, interoperability, explainability, cybersecurity, privacy, bias and the limits of artificial intelligence (AI).

Learning Objectives

On completing this chapter, the learner should be able to:

1. Explain the relationship between BIM-linked material data, digital twins and asset information management.

2. Design a material-data structure that supports selection, traceability and quality control.

3. Interpret sensor, laboratory and computer-vision data for construction-material applications.

4. Select appropriate statistical and machine-learning approaches for prediction and anomaly detection.

5. Evaluate data quality, uncertainty, model performance and explainability.

6. Propose a governed and cybersecurity-aware digital materials workflow.

7. Critically assess the limitations and ethical implications of AI-assisted engineering decisions.

1. From Material Records to Digital Materials Engineering

1.1 The material information chain

A material information chain links four related elements:

• Physical material: the actual batch, component or assembly.
• Digital identity: a unique identifier, such as a batch, heat, pallet, element or asset ID.
• Material attributes: composition, grade, source, certificates, test results and environmental data.
• Events and observations: production, delivery, placement, inspection, exposure, repair and failure events.

For example, a concrete placement record may link a BIM element to a mix-design ID, supplier batch, delivery ticket, truck arrival time, slump and temperature measurements, cube specimens, curing conditions and later durability observations. The purpose is not to collect every possible variable, but to preserve information that changes an engineering decision.

1.2 BIM-linked material information

BIM provides a structured representation of building or infrastructure components. Material information should be linked to objects through stable identifiers rather than through unstructured text alone. A useful object may contain:

• classification and material name;
• manufacturer, product and batch identifiers;
• specification and intended performance;
• geometry, location and quantity;
• declared environmental indicators;
• inspection and test records;
• installation constraints and safety information;
• inspection, repair and end-of-life requirements.

Object properties should distinguish specified, measured, estimated and derived values. A measured compressive strength, for instance, should not be stored as if it were the specified characteristic strength. Units, test age, test method, specimen type and uncertainty are essential metadata.

Interoperability is a major concern. Open exchange formats and common data environments can reduce dependence on proprietary systems, but exchange is not equivalent to semantic interoperability. Two systems may exchange the word “strength” while referring to different ages, units or test procedures. A project therefore requires a data dictionary, naming convention, classification system, responsibility matrix and rules for version control.

1.3 Digital twins

A digital twin is a maintained digital representation of a physical asset that is connected to the asset through data flows and can support analysis or decisions. A static BIM model is not necessarily a digital twin. A twin generally includes:

1. a physical asset or process;

2. a digital representation;

3. connections for data acquisition and updating;

4. analytical or simulation capability;

5. a defined operational purpose.

A materials digital twin may represent the changing condition of concrete, steel, pavement or composite materials. It can combine design information, sensor observations, inspection images, environmental exposure and deterioration models. The twin is useful only within its validity domain. It does not directly observe every internal mechanism, and modelled condition remains an inference when measurements are sparse.

2. Data Architecture for Materials Engineering

2.1 Data types and provenance

Construction-materials data commonly include:

• Design data: mix proportions, target performance, grades and tolerances.
• Supply-chain data: source, certificates, batch, transport and storage.
• Laboratory data: strength, permeability, chemical composition, particle size and durability tests.
• Field data: temperature, maturity, moisture, vibration, placement and curing.
• Inspection data: photographs, ultrasonic measurements, rebound values, corrosion potentials and defect maps.
• Operational data: loads, environmental exposure, deformations and maintenance history.
• Contextual data: weather, location, responsible organisation and timestamps.

Every value should have provenance: who produced it, when, with what instrument or procedure, and whether it was directly measured, transcribed, transformed or predicted. Provenance permits audit and helps distinguish a sensor fault from genuine material change.

2.2 Data quality dimensions

Data quality is multidimensional:

• Accuracy: closeness to the accepted or true value.
• Precision: repeatability of measurements.
• Completeness: proportion of required fields or events available.
• Consistency: absence of contradictory values across systems.
• Timeliness: suitability for the decision’s time scale.
• Validity: conformance to units, ranges and allowed formats.
• Uniqueness: absence of duplicate records.
• Traceability: ability to identify origin and transformation.

A complete dataset may still be inaccurate; a highly accurate laboratory result may be too late to control fresh concrete placement. Data-quality requirements must therefore be defined against a decision, not treated as an abstract score.

2.3 Data governance

A professional data-governance arrangement assigns ownership and stewardship. It specifies who may create, approve, alter and use records; how long records are retained; which data are authoritative; and how changes are audited. Validation rules should identify impossible values, such as negative curing duration or a temperature outside instrument capability. However, automated cleaning must not silently delete unusual but genuine observations. Suspect values should be quarantined, reviewed and retained with their status.

3. Sensors and Digital Observation of Materials

Sensors convert physical conditions into digital observations. Typical applications include:

• embedded temperature and relative-humidity sensors for concrete curing;
• strain gauges and fibre-optic sensors for deformation;
• corrosion and half-cell measurements for reinforced concrete;
• moisture sensors in timber and building envelopes;
• acoustic emission and ultrasonic systems for cracking or damage;
• weigh systems and RFID tags for material movement;
• environmental sensors for chloride, carbon dioxide, temperature and wetting cycles.

A sensor system requires more than hardware. It requires calibration, installation records, sampling design, time synchronisation, communications, power management, data storage and maintenance. Sensor drift, loss of connectivity, electromagnetic interference, poor coupling and environmental damage can produce misleading records.

3.1 Signal processing and anomaly detection

Raw sensor data often contain noise, missing observations and outliers. A moving average can reduce short-term random variation, but it may also hide a sudden failure. Filtering should therefore be selected according to the physical process and decision purpose. A sudden jump in concrete temperature may represent hydration, a sensor displacement or a data-transmission error; context is required.

An anomaly is not automatically a defect. It is an observation or pattern inconsistent with an established baseline. Baselines should account for location, season, loading and construction stage. Simple control charts may be more transparent and robust than complex AI models when data are limited.

3.2 Concrete maturity as an example

Temperature monitoring can support estimation of strength development through a maturity relationship, provided the concrete mixture and curing conditions are adequately characterised. A common temperature-time maturity index is:

\[

M=\sum_{i=1}^{n}(T_i-T_0)\Delta t_i

\]

where \(M\) is the maturity index, \(T_i\) is the average concrete temperature during interval \(i\), \(T_0\) is the datum temperature below which the assumed strength-development contribution is zero, and \(\Delta t_i\) is the interval duration. Temperature and datum temperature must use the same scale, commonly degrees Celsius for differences, and time units must be consistent.

The equation assumes that the selected maturity function adequately represents the mixture and that temperature history is a meaningful predictor of strength. It should not be used to release formwork or open a structure without calibration, verification and compliance with applicable standards. A maturity estimate is not a direct strength measurement; it is a model-based inference.

4. Computer Vision for Materials and Construction

Computer vision uses images or video to classify, measure or locate features. Applications include:

• aggregate grading and shape assessment;
• identification of surface cracks, spalling, honeycombing and staining;
• verification of product labels and protective equipment;
• monitoring of stockpiles and material quantities;
• assessment of surface finish and coating continuity;
• detection of construction-stage changes.

A typical workflow includes image acquisition, calibration, preprocessing, feature extraction or segmentation, classification, measurement and engineering review. Illumination, camera angle, scale, occlusion, surface contamination and image resolution strongly affect performance.

A crack-detection model trained on clean laboratory images may fail on shaded, dirty or textured field surfaces. Performance should be reported separately for relevant conditions, not only as one overall accuracy value. False negatives can be more serious than false positives in safety-related inspection, but excessive false alarms can cause inspection fatigue and loss of trust.

Computer vision does not directly measure structural significance. A visible crack’s importance depends on width, depth, orientation, location, reinforcement, loading, exposure and change over time. Vision should therefore support, not replace, engineering diagnosis.

5. Analytics and AI for Material Decisions

5.1 Descriptive, diagnostic, predictive and prescriptive analytics

Four levels are useful:

• Descriptive analytics: What happened? Examples are batch rejection rates and moisture histories.
• Diagnostic analytics: Why did it happen? Examples are association of low strength with curing temperature or transport delay.
• Predictive analytics: What is likely to happen? Examples are predicted strength, corrosion initiation or remaining service performance.
• Prescriptive analytics: What action should be taken? Examples are changing curing practice, increasing inspection or selecting a replacement material.

Each level requires assumptions. Prescriptive outputs should not be treated as objective truths because they depend on costs, constraints, risk tolerance and the quality of predicted outcomes.

5.2 Statistical and machine-learning models

Regression models can estimate a response from explanatory variables. For a linear model:

\[

\hat{y}=\beta_0+\beta_1x_1+\cdots+\beta_px_p

\]

where \(\hat{y}\) is the predicted response, \(x_j\) are explanatory variables, \(\beta_j\) are fitted coefficients and \(\beta_0\) is the intercept. The model assumes that the chosen variables and functional form are suitable; ordinary least-squares inference additionally requires suitable independence, error structure and consideration of collinearity. A high fit to historical data does not establish causation.

Classification models estimate categories such as acceptable or requiring inspection. Clustering can reveal groups without labelled outcomes, while time-series models use temporal structure. Neural networks can model complex relationships but usually require substantial, representative data and careful validation. Tree-based models are often effective for tabular materials data and can provide more accessible explanations than deep networks, although interpretability is not guaranteed.

5.3 Model evaluation

Data should be separated into training, validation and test sets, preferably by project, time period or material source where leakage is possible. Randomly splitting highly similar observations can produce an unrealistically optimistic result. For regression, mean absolute error is:

\[

MAE=\frac{1}{n}\sum_{i=1}^{n}|y_i-\hat{y}_i|

\]

where \(y_i\) is the observed value, \(\hat{y}_i\) is the prediction and \(n\) is the number of observations. MAE has the same units as the response and is readily interpretable. Root mean squared error gives greater weight to large errors, while a coefficient of determination may be misleading when the application has a narrow range or systematic bias.

For classification, sensitivity measures the proportion of actual positive cases detected, while specificity measures the proportion of actual negative cases correctly rejected. Their relative importance depends on consequences. A defect-screening system may prioritise sensitivity, followed by human confirmation of flagged locations.

Uncertainty should accompany predictions. A point estimate such as predicted chloride concentration is incomplete without an indication of confidence, prediction interval or model applicability. Uncertainty arises from measurement error, sampling, model form, parameter estimation and conditions not represented in training data.

5.4 Explainability and human oversight

Explainability may be:

• Intrinsic: the model structure is understandable, as in a simple decision rule.
• Post hoc: an explanation is generated after a complex model produces an output.
• Global: explanation of general model behaviour.
• Local: explanation of one prediction.

Feature importance does not prove causation. An algorithm may associate low strength with a supplier ID because that identifier indirectly captures an unrecorded change in mixture or curing practice. Engineers should ask whether the relationship is physically plausible, stable across projects and consistent with independent evidence.

Safety-critical decisions require a defined human-override process, documented competence, escalation criteria and an audit trail. Human oversight must be meaningful; simply requiring a person to click “approve” is not adequate control.

6. Material Selection Using Digital Methods

Digital material selection can combine performance, cost, carbon, availability, durability and constructability. A multi-criteria decision model may represent an overall score as:

\[

S=\sum_{j=1}^{m}w_jz_j

\]

where \(z_j\) is a normalised criterion value and \(w_j\) is its weight, with weights commonly constrained to sum to one. The method assumes that criteria can be normalised comparably and that additive trade-offs are acceptable. It should not conceal mandatory constraints: a material failing fire, structural, health or durability requirements must be rejected regardless of its composite score.

A robust selection process therefore has two stages:

1. eliminate options that fail codes, safety requirements, exposure limits, supply constraints or verified performance thresholds;

2. rank feasible options using transparent criteria and sensitivity analysis.

Sensitivity analysis tests whether the preferred material changes when weights, prices, service-life assumptions or carbon factors vary. This is important because environmental declarations, recycled-content claims and service-life predictions may have different boundaries and uncertainty.

7. Traceability Across the Supply Chain

Traceability links material identity and status from source to installed location. Technologies include barcodes, QR codes, RFID, laboratory information systems, GPS records and distributed ledgers. A distributed ledger can improve tamper evidence among organisations, but it cannot guarantee that the original entry was truthful. “Garbage in” remains a fundamental limitation.

A traceability system should define:

• the unit of identification, such as batch, heat, roll or component;
• custody-transfer events;
• required evidence at each stage;
• acceptance and quarantine states;
• links between physical labels and BIM objects;
• procedures for damaged, substituted or rejected materials.

Traceability supports targeted recalls and investigation. It also reduces unnecessary replacement because affected batches can be isolated rather than treating an entire project as suspect.

8. A Digital Twin Workflow for Quality and Maintenance

A practical workflow contains the following stages:

1. Define decisions and consequences. Identify what action the system must support and the risk of error.

2. Specify information requirements. Define required attributes, tolerances, timing, units and responsible parties.

3. Create identities and linkages. Connect batches, products and installed elements to BIM objects.

4. Acquire and validate data. Calibrate instruments, verify records and document provenance.

5. Integrate models. Combine deterministic engineering models, statistical analysis and inspection evidence.

6. Evaluate and communicate uncertainty. Report confidence, applicability and data gaps.

7. Act and record. Document decisions, approvals, interventions and outcomes.

8. Learn and update. Compare predictions with observations and revise the model under controlled governance.

This is a closed learning loop, but it must avoid uncontrolled self-modification. Changes to a model used for acceptance or maintenance should be versioned, validated and approved.

9. Illustrative Case: Digital Quality and Maintenance Management for a Coastal Concrete Bridge

Background

The following is an illustrative case, not a report of a particular real project. A reinforced-concrete bridge in a chloride-exposed coastal environment has repeated records of inconsistent curing and increasing surface cracking. The owner wants to reduce unnecessary inspections while ensuring that deterioration is detected early.

Problem

Construction records are stored as paper certificates, inspection images have inconsistent location descriptions, and several embedded sensors have missing data. The maintenance team cannot reliably connect a crack photograph to a concrete batch or exposure history. A vendor proposes an AI system that predicts corrosion risk from images and environmental data.

Analysis

The project team first defines the decision: identify zones requiring priority inspection and estimate whether intervention should be scheduled. It does not attempt to let AI authorise structural repairs. A BIM model is cleaned and each deck segment, pier and repair zone receives a stable identifier. Concrete batch, placement date, curing record, strength results and supplier information are linked to these identifiers.

Sensor data are checked for time drift, impossible values and periods of communication loss. Missing data are labelled rather than silently interpolated. Drone images are collected with calibrated scale references and lighting notes. Existing cracks are mapped by location, width and change over time. A small, reviewed image set is created for model development, with separate validation by bridge zone to reduce leakage between nearly identical images.

A baseline deterioration model uses exposure, cover, crack history, half-cell observations and inspection intervals. A computer-vision model flags probable cracks, but an engineer verifies the location and significance. Predictions are reported with uncertainty and a rule prevents high-priority decisions when essential cover or exposure data are absent.

Approach and solution

The owner implements a governed digital twin with three layers:

• an authoritative asset and material register;
• a condition-observation layer containing sensors, tests and images;
• an analytical layer containing deterioration predictions and inspection priorities.

Role-based access, encryption, multifactor authentication and network separation are introduced. A model register records training data, version, performance by condition, limitations and approval status. Maintenance staff receive training on false positives, false negatives and escalation procedures.

Outcome

The system improves traceability from a defect to its construction and exposure history. Inspections become risk-prioritised rather than purely calendar-based, while all safety-significant findings remain subject to professional review. The project also discovers that some apparent deterioration patterns were caused by a sensor installation problem, demonstrating the importance of data-quality investigation before model refinement.

Learning points

The case illustrates that digital twins are socio-technical systems. Their effectiveness depends on identifiers, standards, field practice, data stewardship, engineering models and organisational response. AI is useful for screening and prioritisation, but not a substitute for diagnosis, verification or accountability.

10. Cybersecurity, Ethics and Regulation

Digitally connected material systems create new attack surfaces. An attacker might alter a test certificate, falsify sensor readings, interrupt a supply-chain record or access sensitive facility information. Controls should include least-privilege access, strong authentication, encryption in transit and at rest, secure backups, patch management, network segmentation, logging, incident response and supplier-security requirements.

Cybersecurity is also a safety issue. Integrity and availability can be as important as confidentiality. A bridge operator needs a safe fallback procedure when the digital twin is unavailable or suspected to be compromised.

Ethical issues include worker monitoring, consent, ownership of project data, unequal access to digital tools, automated rejection of suppliers and opaque decisions affecting safety or employment. Data minimisation and purpose limitation should be applied. Organisations should disclose material assumptions and avoid presenting probabilistic outputs as certainties.

Regulatory compliance remains essential. Digital tools do not replace building codes, material standards, contractual inspection duties, professional registration requirements or records obligations. Responsibility for an engineering decision must remain identifiable even when software contributed to it.

11. Limitations and Failure Modes of AI

AI systems can fail through:

• distribution shift: new climates, suppliers, materials or construction methods differ from training data;
• selection bias: only successful or easily measured cases are represented;
• label error: inspection classifications are inconsistent or uncertain;
• data leakage: information unavailable at decision time enters model training;
• confounding: a correlated variable is mistaken for a causal mechanism;
• overfitting: the model memorises projects rather than learning general behaviour;
• automation bias: users accept a system output because it appears objective;
• concept drift: relationships change as materials, practices or environments change;
• adversarial or accidental manipulation: inputs are deliberately or unintentionally altered.

No performance metric removes these risks. Validation should include out-of-project testing, stress tests, review of failure cases, monitoring after deployment and a retirement or retraining rule. Models should be used within a clearly stated domain of applicability.

12. Best-Practice Implementation Framework

An organisation planning digital materials engineering should proceed incrementally:

1. Begin with a high-value, bounded use case such as concrete-batch traceability or image-assisted defect screening.

2. Establish a common data dictionary, identifiers and authoritative sources.

3. Measure baseline performance before automation.

4. Use open, documented interfaces and avoid unnecessary vendor lock-in.

5. Validate sensors and analytical models against independent engineering evidence.

6. Report uncertainty, subgroup performance and known failure conditions.

7. Keep human approval for safety-critical decisions.

8. Protect systems through cybersecurity-by-design and tested fallback procedures.

9. Audit data, model versions, decisions and outcomes.

10. Review benefits using cost, safety, environmental and operational indicators—not novelty or software usage alone.

Contemporary Developments

Current development is moving toward semantically richer digital product passports, automated extraction of information from specifications, physics-informed machine learning, edge computing and multimodal systems that combine text, images, sensor data and BIM. These developments may improve search and prediction, but they increase the need for provenance, access control and evaluation. Large language models can assist with document retrieval, coding and inspection-report drafting; they should not be treated as authoritative sources without retrieval verification and professional review.

Chapter Summary

Digital materials engineering treats construction materials as traceable physical assets connected to structured data, observations and engineering models. BIM provides an information context, while a digital twin adds maintained links to a physical asset and supports analysis or action. Sensors and computer vision expand observation, but calibration, context and data provenance determine whether observations are useful.

Analytics ranges from descriptive reporting to predictive and prescriptive decision support. Statistical and machine-learning models require representative data, appropriate validation and explicit uncertainty. Explainability helps engineers challenge implausible relationships, but feature importance does not establish causation. Digital material selection should first enforce mandatory technical constraints and then use transparent, sensitivity-tested criteria.

Traceability systems can connect supply, installation and maintenance records, although digital records cannot correct false inputs. AI is valuable for screening, prioritisation and pattern detection, but it is vulnerable to bias, distribution shift, leakage, overfitting and automation bias. Cybersecurity, ethics, regulatory compliance, professional accountability and fallback procedures are therefore integral to digital materials practice. The most reliable implementation is governed, incremental, interoperable and centred on well-defined engineering decisions.

Key Terms

BIM: A structured digital representation of built assets and their information.

Digital twin: A maintained digital representation connected to a physical asset through data and analytical processes.

Material provenance: The documented origin, processing and transformation history of material information.

Traceability: The ability to follow a material’s identity, custody, status and installed location.

Data dictionary: A controlled description of data fields, meanings, formats and units.

Data quality: The fitness of data for a defined decision, including accuracy, completeness and timeliness.

Sensor drift: A gradual change in sensor output unrelated to the measured physical quantity.

Computer vision: Computational analysis of images or video to detect, classify or measure features.

Anomaly detection: Identification of observations inconsistent with an established baseline or pattern.

Predictive analytics: Analysis used to estimate future or unobserved outcomes.

Machine learning: Algorithms that infer patterns from data for prediction or classification.

Data leakage: Use of information in model development that would not be available at decision time.

Distribution shift: A change between the data used for model development and deployment conditions.

Explainability: The extent to which a model’s output and reasoning can be understood and examined.

Uncertainty: Lack of complete knowledge about measurements, parameters, models or future conditions.

Model applicability domain: The conditions for which a model has been validated and should be trusted.

Digital product passport: A structured record of a product’s identity, properties, impacts and life-cycle information.

Cybersecurity: Protection of digital systems and data from unauthorised access, alteration or disruption.

Human-in-the-loop: A process in which competent people review, approve or override automated outputs.

Model governance: Controlled management of model development, validation, deployment, monitoring and retirement.

Review Questions

Short Answer Questions

1. Distinguish a static BIM model from a digital twin.

2. Why are provenance and units important in a material database?

3. State two causes of unreliable sensor data.

4. What is data leakage in machine-learning development?

5. Why should computer vision not independently determine the structural significance of a crack?

Long Answer Questions

1. Explain how BIM-linked material data can support quality assurance and maintenance throughout an asset’s life cycle.

2. Discuss the design of a governed digital-twin workflow for a concrete infrastructure asset.

3. Compare descriptive, diagnostic, predictive and prescriptive analytics in construction-materials practice.

4. Explain the principal sources of uncertainty in AI-assisted material and condition assessment.

5. Critically evaluate cybersecurity, ethical and professional-accountability requirements for connected material systems.

Analytical / Application Questions

1. A project has accurate laboratory strength results but incomplete curing records and inconsistent batch identifiers. Propose a data-quality improvement plan before developing a strength-prediction model.

2. Design a decision framework for selecting among three façade materials using durability, embodied carbon, cost and maintainability while ensuring that code and fire requirements are not traded away.

3. An image model reports high overall accuracy but misses cracks on shaded, contaminated surfaces. Explain how you would investigate the failure and decide whether the model is suitable for field deployment.

Multiple Choice Questions with Answers

Question 1

Which feature most clearly distinguishes a digital twin from a static BIM model?

A. The twin contains three-dimensional geometry

B. The twin is connected to the physical asset and maintained for a defined purpose

C. The twin uses proprietary software

D. The twin contains more colours and annotations

Correct Answer: B. The twin is connected to the physical asset and maintained for a defined purpose

Question 2

Which data-quality dimension concerns closeness to the accepted or true value?

A. Completeness

B. Timeliness

C. Accuracy

D. Uniqueness

Correct Answer: C. Accuracy

Question 3

In the maturity relationship M = Σ(Ti − T0)Δti, what does T0 represent?

A. The maximum concrete temperature

B. The datum temperature below which contribution is assumed zero

C. The average ambient temperature

D. The final curing temperature

Correct Answer: B. The datum temperature below which contribution is assumed zero

Question 4

What is the main purpose of separating training, validation and test data?

A. To increase the number of sensors

B. To estimate performance on data not used to fit the model

C. To eliminate all measurement error

D. To guarantee causal relationships

Correct Answer: B. To estimate performance on data not used to fit the model

Question 5

Which statement about feature importance is correct?

A. It always proves causation

B. It is unnecessary for safety-critical models

C. It can indicate associations but does not by itself prove causation

D. It removes distribution shift

Correct Answer: C. It can indicate associations but does not by itself prove causation

Question 6

Which approach is most appropriate in multi-criteria material selection?

A. Rank every option before checking code compliance

B. Reject options failing mandatory requirements before ranking feasible options

C. Use embodied carbon alone

D. Allow a low fire rating to be offset by low cost

Correct Answer: B. Reject options failing mandatory requirements before ranking feasible options

Question 7

What does a false negative represent in a defect-screening system?

A. A sound area incorrectly flagged as defective

B. A defect that the system fails to identify

C. A duplicated image

D. A missing sensor timestamp

Correct Answer: B. A defect that the system fails to identify

Question 8

Which practice best protects the integrity of a connected material-data system?

A. Shared passwords for all project participants

B. Unlogged changes to certificates

C. Role-based access, audit logs and secure backups

D. Deleting unusual sensor readings automatically

Correct Answer: C. Role-based access, audit logs and secure backups

Question 9

Why can a distributed ledger not by itself guarantee trustworthy material records?

A. It cannot store identifiers

B. It cannot correct false information entered at the source

C. It prevents all users from viewing data

D. It cannot record transactions

Correct Answer: B. It cannot correct false information entered at the source

Question 10

What is distribution shift?

A. A change between development data and deployment conditions

B. A change in file format only

C. A method for calibrating a sensor

D. A technique for normalising criteria

Correct Answer: A. A change between development data and deployment conditions