Digital Construction, AI and Safety Technology

Introduction and Context

Construction safety management is increasingly shaped by digital systems that can collect, integrate and interpret information about people, plant, materials, temporary works and changing site conditions. Building Information Modelling (BIM), digital twins, unmanned aerial vehicles (UAVs), wearable devices, Internet of Things (IoT) sensors, computer vision, robotics, geofencing and artificial intelligence (AI) are no longer experimental concepts confined to research laboratories. They are being applied to improve hazard identification, coordination, monitoring, emergency response and organisational learning.

However, digitalisation does not automatically create safer construction. A technically sophisticated system may produce unsafe outcomes if its data are incomplete, poorly governed or misunderstood. Automated alerts may be ignored because of excessive false alarms; computer vision may perform less accurately for some workers or environmental conditions; wearable monitoring may undermine trust; and predictive models may reproduce historical reporting biases. Cybersecurity failures can also affect physical safety when digital systems control access, lifting equipment, temporary works information or emergency communications.

The central professional question is therefore not whether a technology is innovative, but whether it improves risk control within a defensible management system. This requires alignment between technology, work processes, competence, ethical governance and human judgement.

Learning Objectives

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

1. Explain the role of BIM and digital twins in construction safety planning and control.

2. Evaluate the safety applications and limitations of drones, wearables, IoT sensors, computer vision and robotics.

3. Design a governance approach for geofencing and digital safety monitoring.

4. Explain how predictive AI can support, but not replace, professional risk assessment.

5. Assess data quality, cybersecurity, privacy, bias, explainability and human oversight.

6. Propose an integrated digital safety strategy for a construction project.

1. Digital Safety as a Socio-Technical System

A construction site is a complex socio-technical system. Safety outcomes arise from interactions between workers, supervisors, equipment, design decisions, procedures, commercial pressures, environmental conditions and organisational culture. Digital tools form an additional layer in this system. They sense conditions, represent work, communicate information or recommend actions.

A useful classification is based on the function performed:

• Representation: BIM models, geographic information systems and digital twins represent assets, spaces, sequences and status.
• Sensing: IoT devices, environmental monitors, proximity tags and wearables collect observations.
• Observation: Drones and computer vision inspect work areas and behaviours.
• Prediction: AI models estimate the likelihood of events or identify patterns associated with risk.
• Intervention: Geofencing, automated warnings, robotics and access controls influence behaviour or physical operations.
• Learning: Digital records support investigations, trend analysis and continuous improvement.

These functions are related but not interchangeable. A BIM model may show where a crane is intended to operate, but it does not prove that the crane is currently safe. A wearable can detect a fall or heat exposure, but it may not explain why the worker entered a hazardous area. A predictive model can identify a pattern, but it cannot establish legal causation or replace a competent person’s assessment.

Digital safety should consequently be treated as a control cycle:

plan → sense → interpret → decide → intervene → verify → learn

Failure at any stage can reduce safety. For example, inaccurate sensor calibration compromises sensing; an opaque algorithm compromises interpretation; unclear authority compromises decision-making; and lack of follow-up compromises learning.

2. BIM-Based Safety Planning

2.1 BIM foundations

BIM is a structured process for creating, managing and exchanging information about a built asset across its life cycle. A model may contain geometry, properties, relationships, classification, sequencing information and links to documents. For safety management, its value lies less in three-dimensional visualisation alone than in the integration of safety-relevant information with design and construction planning.

Safety applications include:

• identifying fall edges, openings and restricted zones;
• reviewing plant movement and lifting paths;
• coordinating temporary works and access routes;
• checking spatial conflicts between trades;
• planning emergency exits, fire access and assembly areas;
• linking hazards and controls to specific model elements;
• simulating construction sequences and work-at-height activities;
• communicating risks through visual, location-specific information.

A safety-enabled BIM object might include installation sequence, required exclusion zone, inspection status, responsible party and control measures. This transforms a safety plan from a static document into information associated with the work location and activity.

2.2 Four-dimensional and five-dimensional safety planning

4D BIM links model elements to time or construction sequence. It allows safety professionals to examine when hazards emerge, overlap or disappear. For example, a floor opening may be safe during one phase, protected by a temporary cover in another and enclosed by permanent structure later. A 4D review can reveal simultaneous operations that create interaction risks, such as lifting over occupied work areas.

5D BIM commonly adds cost information. Its safety value is indirect but important: it can help compare the cost and programme implications of alternative controls, temporary works and safer methods. The lowest initial cost is not necessarily the lowest whole-life risk.

BIM-based safety planning is strongest when safety personnel participate during design and preconstruction, rather than receiving an information model after decisions have been fixed. Design-for-safety reviews can identify maintenance, access, lifting and construction hazards before work begins.

2.3 BIM limitations

BIM does not solve safety problems automatically. Important limitations include:

• models may omit temporary works, site logistics or informal work practices;
• model information may become obsolete when the site changes;
• different software platforms may interpret data inconsistently;
• responsibility for updating safety information may be unclear;
• visual realism may create false confidence in model accuracy;
• workers may not have access to suitable devices or training.

A model should therefore have an identified information owner, revision status, approval process and field-verification requirement. Critical controls must not depend solely on an unverified model.

3. Digital Twins and Dynamic Safety Management

A digital twin is a digital representation of a physical asset or process that is connected, to an appropriate degree, with current data from the physical system. Unlike a conventional design model, a digital twin is intended to reflect changing conditions and support monitoring, analysis or decision-making.

In construction, the concept can apply to:

• a building under construction;
• a tunnel or excavation;
• a crane and lifting operation;
• a temporary works system;
• a complete project site and its logistics network.

A safety-oriented digital twin may combine BIM geometry, sensor readings, inspection records, weather data, equipment status, work permits and schedule information. It can display whether a monitored area is occupied, whether environmental conditions exceed a threshold or whether a temporary support has been inspected.

The distinction between a model and a twin is one of connection and operational use. A static model describes intended or recorded conditions. A twin incorporates observations and may support feedback to the physical system. The distinction is not absolute; digital twins exist on a spectrum of data connection, update frequency, automation and analytical capability.

3.1 Conditions for a useful digital twin

A digital twin requires:

1. A defined purpose: for example, monitoring excavation movement or coordinating site access.

2. A system boundary: the assets, activities and hazards included must be specified.

3. Reliable identity and location: physical objects must be matched to their digital counterparts.

4. Data connectivity: sensors, inspection systems and models must exchange information.

5. Validation: digital observations must be checked against physical reality.

6. Decision protocols: users must know who receives alerts and what action is required.

A twin that produces dashboards without decision authority or response procedures is an information display, not an effective safety control.

4. Drones, Wearables and IoT Sensing

4.1 Drones and remote inspection

Drones can capture images, video, photogrammetry and thermal information without exposing inspectors to some hazards. Applications include inspection of roofs, façades, towers, stockpiles, earthworks, bridges and inaccessible temporary works. Repeated flights can document progress and identify changes in excavation geometry or material storage.

Safe drone deployment requires consideration of airspace restrictions, weather, battery condition, loss of control, collision risk, privacy, pedestrian exposure and competence of the operator. A drone inspection may reduce exposure for an inspector but introduce risk to workers below. Flight planning, exclusion zones and emergency landing procedures remain necessary.

4.2 Wearable technologies

Wearables include location tags, inertial sensors, physiological monitors, smart helmets, connected safety vests and devices capable of detecting falls or impacts. Potential uses include:

• worker location during emergency response;
• proximity warnings near moving plant;
• detection of falls or prolonged immobility;
• heat-stress and fatigue indicators;
• exposure monitoring for noise, vibration or hazardous substances;
• recording worker observations and near misses.

Wearable data are not equivalent to direct measurements of competence, attention or productivity. A heart-rate increase may reflect heat, exertion, anxiety or an ordinary task. Physiological inference is particularly sensitive and should not be used for disciplinary decisions without strong validation, transparency and due process.

4.3 IoT sensors

IoT systems connect sensors, communication networks, data platforms and user interfaces. Construction safety applications include monitoring:

• dust, gases, noise, temperature and humidity;
• vibration and structural movement;
• water levels and flooding;
• equipment position and operating status;
• access, occupancy and restricted areas;
• lifting loads and temporary works behaviour.

The engineering value of an IoT system depends on sensor selection, placement, calibration, sampling frequency, communication reliability and maintenance. A sensor may be accurate under laboratory conditions but unreliable when covered by dust, exposed to vibration or moved from its intended location.

5. Computer Vision and Robotics

5.1 Computer vision

Computer vision uses cameras and algorithms to classify objects, detect events or estimate states. On construction sites it may identify missing personal protective equipment, people entering exclusion zones, unsafe proximity to equipment, unprotected edges, poor housekeeping or vehicle conflicts.

The performance of a vision system depends on lighting, camera angle, occlusion, weather, image resolution, background complexity and the representativeness of training data. “Detection” is not the same as “understanding.” A camera may detect that a person is near a machine but not know whether the machine is isolated, whether the proximity is authorised or whether the work method makes the condition safe.

Performance should be assessed using a confusion matrix. For a defined event:

• True positive: the system raises an alert and the event is present.
• False positive: the system raises an alert but the event is absent.
• True negative: no alert is raised and the event is absent.
• False negative: no alert is raised although the event is present.

Precision is the proportion of alerts that are correct:

Precision = true positives / (true positives + false positives)

Recall, also called sensitivity, is the proportion of actual events detected:

Recall = true positives / (true positives + false negatives)

These measures apply to a defined dataset and event definition. They are not universal properties of an algorithm. A system may have high recall but poor precision, producing alert fatigue, or high precision but poor recall, missing hazardous events. Threshold selection should reflect the consequence of each error and should be reviewed in operational conditions.

5.2 Robotics and autonomous equipment

Robotics can reduce human exposure to repetitive, heavy, unstable or hazardous tasks. Examples include robotic drilling, surveying, demolition, bricklaying assistance, autonomous haulage and remotely operated equipment. Safety benefits arise when the robot replaces exposure rather than merely adding another machine to a crowded work area.

Robotic systems require risk assessment for unexpected movement, loss of communication, software failure, maintenance access, human-machine interaction and mode confusion. A worker must be able to identify the robot’s operating state and understand how to stop it. Autonomous operation should have defined boundaries, fail-safe behaviour and a tested manual recovery method.

6. Geofencing and Proximity Control

Geofencing creates a virtual boundary using geographic coordinates, local positioning, radio identification or other technologies. It may generate warnings, restrict equipment movement or trigger access controls when a person or asset enters a defined zone.

Applications include separating pedestrians from vehicles, restricting access to lifting zones, protecting excavation edges and warning workers near overhead hazards. A robust system distinguishes between:

• the hazard zone, where exposure may occur;
• the buffer zone, allowing time for detection and response;
• the safe zone, where normal operation is permitted.

Geofencing should not be treated as a substitute for physical segregation, barriers, competent supervision or equipment controls. Positioning errors, signal loss, latency, battery failure and incorrect zone configuration can create dangerous gaps. The system must specify what happens when data are unavailable: normally, a critical control should fail to a safe state or require a conservative manual procedure.

7. Predictive AI for Construction Safety

Predictive AI uses historical and current data to classify conditions, estimate probabilities, prioritise inspections or identify patterns associated with incidents. Inputs may include near-miss reports, inspection findings, weather, schedule pressure, subcontractor interfaces, workforce characteristics, equipment data and task type.

A generic risk score may be represented as:

R = P(E | X) × C

where:

• R is a prioritisation score;
• P(E | X) is the model-estimated probability of event E given data X;
• C is the consequence rating or consequence estimate.

This expression is useful as a conceptual decision aid, but it is not a universal law of risk. It assumes that probability and consequence can be meaningfully estimated and combined for the decision being made. In practice, consequence categories may be ordinal rather than numerical, and probabilities may be poorly calibrated. The score must therefore be interpreted with professional judgement and not presented as an objective measure of safety.

Predictive systems can support:

• prioritisation of inspections;
• identification of recurring hazard combinations;
• allocation of safety resources;
• early warning of schedule or environmental conditions;
• analysis of near misses and leading indicators.

They can also cause harm. Incident data are often under-reported, and reporting may be more complete in teams with stronger safety cultures. A model could therefore label transparent teams as higher risk than teams that conceal events. Historical data may reflect unequal supervision, biased enforcement or unequal access to protective equipment. Predictive outputs should not be used to stigmatise workers, automatically penalise subcontractors or justify reduced attention to apparently low-risk groups.

7.1 Model evaluation and explainability

A responsible predictive system requires separation of training, validation and testing data, monitoring for changes in site conditions and assessment of calibration. A calibrated model is one in which predicted probabilities correspond reasonably to observed frequencies over an appropriate population. Performance should be checked across relevant groups, tasks, shifts and environmental conditions.

Explainability has several levels:

• global explanation: the general factors influencing model behaviour;
• local explanation: why a particular case received an alert or score;
• procedural explanation: who designed, validated and approved the system;
• action explanation: what a user should do in response to the output.

An explanation is useful only if it helps a competent person challenge, verify or act on the result. Complex mathematical transparency without operational meaning is insufficient.

8. Data Quality and Information Governance

Digital safety decisions are only as reliable as the information chain that supports them. Important dimensions of data quality include:

• accuracy: closeness to the actual condition;
• completeness: whether required fields and observations are present;
• timeliness: whether information reflects current conditions;
• consistency: whether definitions and units agree across systems;
• provenance: whether the source and processing history are known;
• validity: whether the data conform to required formats and rules;
• availability: whether authorised users can access information when needed.

A practical data governance plan should define data owners, collection methods, retention periods, access rights, validation checks, correction procedures and audit trails. Critical data should have a status such as draft, verified, superseded or rejected. Sensor readings should include time, location, unit, calibration status and quality flags.

Interoperability is a major issue. BIM, scheduling, enterprise, sensor and incident systems may use different identifiers and data structures. Common information requirements and agreed naming conventions reduce ambiguity. Data minimisation is also important: collecting more information than necessary increases privacy, security and management burdens.

9. Cybersecurity and Physical Safety

Construction organisations are exposed to phishing, ransomware, credential theft, malicious software, unauthorised access and manipulation of operational data. A cyber incident can become a safety incident if it disables alarms, alters a model, falsifies inspection status, unlocks restricted access or interrupts emergency communication.

A safety-conscious cybersecurity programme should include:

• asset and system inventories;
• role-based access and least privilege;
• multi-factor authentication for critical systems;
• secure configuration and timely patching;
• network segmentation between business and operational systems;
• encrypted communication where appropriate;
• tested backups and recovery procedures;
• supplier and software-security assessment;
• incident reporting and exercise-based response planning.

Security controls must be proportionate. Excessive access restrictions can delay urgent work, while weak controls expose critical systems. Change management is essential: a software update, sensor replacement or model revision may alter safety performance and must be tested before deployment.

10. Privacy, Ethics and Human Oversight

Digital monitoring can collect identifiable location, image, voice, health, biometric and behavioural information. Organisations should establish a lawful and legitimate purpose, provide clear notice, limit collection, restrict access, retain information only as long as justified and provide routes for correction or challenge. Privacy impact assessment is appropriate where monitoring is systematic, intrusive or likely to affect workers significantly.

Consent alone may be inadequate in employment settings because workers may not feel free to refuse. Consultation with workers and representatives is therefore necessary. Monitoring should focus on hazard control rather than unnecessary surveillance. Video and wearable data should not be repurposed for productivity measurement without a separate, transparent justification.

Human oversight means more than placing a person nominally in the process. The human decision-maker must have competence, authority, sufficient time, access to relevant evidence and the ability to override the system. Automated decisions affecting access, discipline, work allocation or emergency action should be reviewable and contestable.

Bias can enter through unrepresentative training data, unequal sensor coverage, language barriers, different PPE or clothing, lighting conditions and historical organisational practices. Testing should include diverse workers, tasks, body types, weather conditions and site layouts. Where performance differs materially, deployment should be limited, corrected or discontinued.

11. Implementation Framework for Digital Safety Technology

A construction organisation can evaluate a proposed technology through the following stages.

Stage 1: Define the safety problem

State the hazard, exposed population, existing controls, desired outcome and decision to be improved. “Use AI to improve safety” is not an adequate problem definition.

Stage 2: Examine the baseline

Assess current incident data, inspection quality, work processes, workforce capability and existing controls. A digital tool should improve a known weakness rather than create a parallel reporting burden.

Stage 3: Select the least intrusive effective technology

Compare engineering controls, physical segregation, procedural measures and digital options. Technology should complement the hierarchy of controls, not be used to shift responsibility onto workers.

Stage 4: Validate through a controlled pilot

Define success measures, test under realistic conditions and record false positives, false negatives, downtime, user acceptance and unintended consequences. Include workers and safety professionals in evaluation.

Stage 5: Integrate with response procedures

Specify alert recipients, response times, escalation routes, verification duties and records. Every alert should have an operational meaning.

Stage 6: Govern and audit

Review performance, privacy, cybersecurity, model drift, accessibility, data quality and changes in work conditions. Stop or modify the system when it is unreliable or harmful.

Useful performance indicators include response time to verified alerts, proportion of critical hazards closed, sensor availability, false-alert burden, worker participation, audit findings and reduction in exposure. Incident reduction alone is insufficient because incidents are relatively infrequent and may be affected by reporting behaviour.

12. Illustrative Case: Integrated Digital Safety on a Metro Station Project

Background

The following is an illustrative case based on a hypothetical underground metro station project. The project involved deep excavation, tower cranes, heavy vehicle movements, multiple subcontractors and changing work fronts.

Problem

Near misses involved pedestrians entering delivery routes, incomplete edge protection and conflicting lifting activities. Existing inspections were paper-based and often recorded conditions several hours after they occurred. The project proposed a combined BIM, IoT, computer-vision and geofencing system.

Analysis

The project team first mapped hazards to work stages. BIM was used to identify excavation edges, crane swing zones, access routes and temporary openings. A 4D review showed that two subcontractors planned overlapping lifting operations. Sensors were proposed for excavation movement and environmental conditions. Cameras were limited to vehicle-pedestrian interfaces and major access points. Wearable location tags were offered for workers entering high-risk zones, with a policy prohibiting use for productivity scoring.

The team identified several risks. The vision system performed poorly at night and during dust events. Positioning tags sometimes lost signal inside the excavation. Historical near-miss data were concentrated in areas with better supervision and could not be treated as an unbiased prediction dataset. Cybersecurity review found that contractor accounts had excessive permissions.

Approach and solution

The project adopted the following controls:

1. Physical barriers and controlled gates were retained as primary controls; geofencing provided supplementary warnings.

2. BIM safety zones were linked to the current work package and required field verification before activation.

3. Sensors were calibrated, assigned unique identifiers and checked during scheduled inspections.

4. Computer-vision alerts were sent first to a trained safety observer for verification rather than directly imposing disciplinary action.

5. Critical alerts had manual backup procedures for network or battery failure.

6. Access was restricted using role-based permissions and multi-factor authentication.

7. Workers were informed about data collection, retention and access, and a worker committee reviewed the monitoring policy.

8. The predictive component prioritised inspection locations but did not determine individual worker risk.

Outcome

The project reported faster verification of vehicle-interface hazards and improved coordination of lifting activities. It also discovered that several apparent “unsafe entries” were authorised work requiring the model’s zone definitions to be corrected. The project therefore treated user feedback and false positives as system-improvement data rather than evidence of worker non-compliance.

Learning points

The case demonstrates that digital safety is most effective when integrated with physical controls, competent supervision and clear governance. It also shows why data quality, fail-safe arrangements, worker participation and the ability to challenge automated outputs are essential.

13. Contemporary Developments and Future Direction

Emerging developments include edge AI, which processes data near the sensor to reduce latency and connectivity dependence; synthetic data for testing computer-vision systems; multimodal systems combining text, images and sensor streams; robotic inspection; and digital permits linked to spatial models. Large language models may assist with searching safety requirements, drafting inspection summaries or identifying inconsistencies, but their outputs require verification because they can produce plausible but incorrect statements.

Future research should examine long-term safety outcomes rather than adoption rates alone. Important questions concern how digital systems affect safety culture, worker autonomy, reporting behaviour, inequality and professional competence. Research methods should combine technical performance testing with field observation, human-factors analysis and participatory evaluation.

Best-Practice Principles

1. Begin with a clearly defined safety problem.

2. Use the hierarchy of controls; do not substitute monitoring for elimination or engineering control.

3. Maintain an authoritative, current source of safety information.

4. Verify models and sensor data against field conditions.

5. Design for failure, including loss of power, connectivity, positioning and software.

6. Measure false positives and false negatives in realistic environments.

7. Protect privacy through purpose limitation and data minimisation.

8. Secure systems as part of physical safety management.

9. Test for bias across workers, tasks and conditions.

10. Ensure competent human oversight, appeal and override.

11. Involve workers in design, testing and governance.

12. Audit performance continuously and retire ineffective tools.

Chapter Summary

Digital construction technologies can strengthen safety planning, monitoring and learning when they are embedded in a socio-technical risk-management system. BIM supports spatial and sequence-based planning, while digital twins connect representations to changing physical conditions. Drones reduce exposure during some inspections; wearables and IoT sensors provide location, environmental and equipment data; computer vision detects selected events; robotics can remove people from hazardous tasks; and geofencing supplements physical separation.

Predictive AI can prioritise inspections and identify patterns, but its outputs are probabilistic and dependent on data quality. Precision, recall and calibration must be evaluated in realistic conditions. Data incompleteness, model drift, false alarms and historical reporting bias can produce unsafe or unfair decisions. Cybersecurity is a safety issue because digital compromise can affect physical operations. Privacy, transparency, worker participation, explainability and human oversight are essential ethical and managerial requirements.

The most defensible implementation approach defines a genuine safety problem, evaluates the baseline, pilots the least intrusive effective technology, integrates alerts with response procedures and audits technical and human performance. Digital tools should augment competent professionals and workers, not obscure responsibility or replace judgement.

Key Terms

BIM: A structured process for creating and managing information about a built asset.

4D BIM: BIM linked to time or construction sequence.

Digital twin: A connected digital representation of a physical asset or process that reflects relevant operational data.

IoT sensing: Networked sensing in which devices collect and exchange data.

Wearable: A body-worn digital device that collects, communicates or displays information.

Computer vision: Algorithmic analysis of images or video to detect objects, events or conditions.

Geofencing: Creation of a virtual geographic boundary that triggers information or controls.

Predictive AI: AI used to estimate future events, probabilities or priorities from data.

Precision: The proportion of system alerts that are correct.

Recall: The proportion of actual defined events detected by a system.

False positive: An alert raised when the defined event is absent.

False negative: Failure to raise an alert when the defined event is present.

Explainability: The ability to communicate how and why a system produced an output.

Model drift: Decline in model performance as conditions or data change.

Data provenance: Information about the origin and processing history of data.

Data minimisation: Limiting collection to information necessary for a defined purpose.

Human oversight: Competent human review, intervention and accountability in an automated process.

Alert fatigue: Reduced attention caused by excessive or unreliable alerts.

Socio-technical system: A system formed by interacting social, organisational and technical elements.

Fail-safe design: Design that moves toward a safe condition when specified failures occur.

Review Questions

Short Answer Questions

1. Distinguish between a BIM model and a digital twin.

2. State two construction safety applications of computer vision.

3. Define a false negative in safety monitoring.

4. Why is cybersecurity relevant to physical construction safety?

5. Identify two privacy risks associated with wearable devices.

Long Answer Questions

1. Explain how 4D BIM can improve the management of changing construction hazards, and discuss its limitations.

2. Evaluate the safety benefits and risks of combining IoT sensors, wearables and geofencing on a large project.

3. Discuss why predictive AI should support, rather than replace, professional safety judgement.

4. Explain the roles of data quality, explainability and human oversight in responsible digital safety management.

5. Develop an implementation framework for introducing computer vision on a live construction site.

Analytical / Application Questions

1. A computer-vision system generates 120 alerts in one week. Of these, 90 are confirmed hazards; inspection records show that 100 hazards actually occurred. Calculate its precision and recall, and explain the operational implications.

2. A project proposes location tracking for all workers to reduce vehicle-pedestrian collisions. Critically assess the proposal using the hierarchy of controls, privacy principles, cybersecurity and worker participation.

3. Design a digital safety architecture for a deep-excavation project using BIM, sensors, drones and predictive AI. Specify data sources, control decisions, failure arrangements and human responsibilities.

Multiple Choice Questions with Answers

Question 1

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

A. Use of three-dimensional geometry

B. Connection to relevant current data from the physical asset or process

C. Inclusion of construction cost information

D. Ability to print drawings

Correct Answer: B. Connection to relevant current data from the physical asset or process

Question 2

What is the principal safety value of 4D BIM?

A. It automatically eliminates all site hazards

B. It links model elements to construction time and sequence

C. It replaces the need for site supervision

D. It measures workers’ physiological condition

Correct Answer: B. It links model elements to construction time and sequence

Question 3

A computer-vision system produces 80 true positives and 20 false positives. What is its precision?

A. 20%

B. 50%

C. 80%

D. 100%

Correct Answer: C. 80%

Question 4

Which situation is a false negative?

A. An alert is raised for a hazard that is not present

B. A hazard is present but the system raises no alert

C. A hazard is present and correctly detected

D. No hazard is present and no alert is raised

Correct Answer: B. A hazard is present but the system raises no alert

Question 5

Why should geofencing generally supplement rather than replace physical segregation?

A. Geofencing cannot display any information

B. Positioning, communication and configuration failures can occur

C. Physical barriers are always less expensive

D. Geofencing is unrelated to vehicle movements

Correct Answer: B. Positioning, communication and configuration failures can occur

Question 6

Which is an example of data provenance?

A. The colour of a dashboard

B. The origin and processing history of a sensor reading

C. The number of workers on a site

D. The consequence rating of an incident

Correct Answer: B. The origin and processing history of a sensor reading

Question 7

What is a major risk of using historical incident reports in predictive AI?

A. They always contain too much information

B. They may reflect reporting and supervision biases

C. They cannot be stored digitally

D. They eliminate the need for validation

Correct Answer: B. They may reflect reporting and supervision biases

Question 8

Which practice best supports privacy-conscious worker monitoring?

A. Collecting all available biometric data indefinitely

B. Using data for unrelated productivity scoring without notice

C. Limiting collection to a defined and justified safety purpose

D. Allowing every contractor unrestricted access

Correct Answer: C. Limiting collection to a defined and justified safety purpose

Question 9

What does human oversight require in an automated safety system?

A. A person is named in the procedure but cannot intervene

B. Competent personnel can review, challenge and override outputs

C. All decisions are made automatically

D. Workers are excluded from system design

Correct Answer: B. Competent personnel can review, challenge and override outputs

Question 10

Which control is most appropriate when a critical digital alert system loses communication?

A. Ignore the loss because the system may recover

B. Continue normal operation without notification

C. Use a predefined conservative manual or fail-safe procedure

D. Delete the affected data

Correct Answer: C. Use a predefined conservative manual or fail-safe procedure