Industrial and Societal IoT Applications

Introduction

The Internet of Things (IoT) has developed from a technology for connecting individual sensors and devices into an infrastructure for observing, analysing and influencing physical systems. Industrial and societal IoT applications extend this idea to factories, energy networks, transport systems, hospitals, farms, buildings, cities and ecosystems. Their value does not arise merely from connectivity. It arises when trustworthy data is converted into decisions, coordinated action and measurable improvement.

IoT deployments in these domains are cyber-physical systems. Sensors observe physical conditions; communication networks transfer data; edge and cloud platforms process information; analytics and artificial intelligence identify patterns; and actuators or human operators implement interventions. The resulting systems must satisfy requirements that are more demanding than those of consumer IoT, including safety, availability, cybersecurity, interoperability, privacy, maintainability and regulatory compliance.

A critical evaluation therefore asks more than whether a device can be connected. It asks whether the architecture is appropriate, whether the data is reliable, whether the intervention improves an outcome, who bears the risks, and whether the system remains economically and socially sustainable at scale.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Explain reference architectures for industrial and societal IoT.

2. Evaluate IoT applications in manufacturing, energy, transport, healthcare, agriculture, buildings, cities and environmental monitoring.

3. Select meaningful technical, economic, operational, environmental and social outcome metrics.

4. Analyse a deployment using reliability, lifecycle, cybersecurity and governance considerations.

5. Distinguish promising applications from projects that merely generate data without decision value.

6. Critically assess the roles of edge computing, artificial intelligence, digital twins and interoperable standards.

1. From Connected Devices to Outcome-Oriented IoT

An application-oriented IoT system can be represented as a sequence of capabilities:

Physical process → sensing and identification → communication → edge processing → platform and data management → analytics and decision → actuation or human response → measured outcome.

This sequence is not necessarily linear. Feedback connects decisions to the physical process, while operational data is used to improve models. A useful architecture commonly contains the following layers:

• Physical and device layer: sensors, actuators, machines, meters, vehicles and identification technologies.
• Connectivity layer: industrial Ethernet, Wi-Fi, cellular networks, LPWAN, fieldbus systems, satellite links and short-range protocols.
• Edge and control layer: gateways, programmable logic controllers, local databases, real-time processing and safety functions.
• Platform layer: device management, data ingestion, storage, digital twins, identity management and APIs.
• Application layer: dashboards, optimization, predictive maintenance, clinical support, traffic management and environmental services.
• Governance layer: security, privacy, safety, compliance, data ownership, service-level agreements and accountability.

The architecture should be matched to application criticality. A temperature dashboard can tolerate seconds or minutes of latency; an industrial emergency shutdown cannot safely depend on a distant cloud service. Consequently, safety-critical and time-critical control remains local, while cloud services are useful for fleet-level analysis, historical trends and model training.

Important terminology

An asset is a physical or logical entity being monitored or controlled. A digital twin is a dynamic digital representation of an asset or process, linked to data and models over its lifecycle. Telemetry is the remote measurement and transmission of operational data. Operational technology (OT) controls or monitors physical processes, whereas information technology (IT) primarily manages information and enterprise computing. Interoperability is the ability of systems to exchange and meaningfully use information. Observability refers to the ability to infer system state from available measurements. Actuation is the deliberate physical change produced by a control device.

2. Evaluation Frameworks and Outcome Metrics

A rigorous evaluation begins with a baseline. The baseline may be historical performance, a control group, an engineering model or a regulatory threshold. Without a baseline, a reduction in energy use or downtime cannot confidently be attributed to IoT.

2.1 Technical and operational metrics

Common metrics include availability, latency, packet-delivery success, sensor accuracy, false-alarm rate, mean time between failures (MTBF), mean time to repair (MTTR), throughput, process capability and service coverage. Availability is often approximated as

\[

A=\frac{\text{MTBF}}{\text{MTBF}+\text{MTTR}},

\]

where MTBF and MTTR use the same time units. This relationship assumes a repairable system with stable average failure and repair behaviour and excludes planned maintenance unless the measurement definition includes it. For systems with redundant components, availability must be derived from the actual series-parallel architecture rather than applying this expression blindly.

For a binary classifier used in predictive maintenance, precision and recall are defined as

\[

\text{Precision}=\frac{TP}{TP+FP}, \qquad

\text{Recall}=\frac{TP}{TP+FN},

\]

where TP, FP and FN denote true positives, false positives and false negatives. Precision matters when unnecessary maintenance is expensive; recall matters when missing a dangerous failure is unacceptable. The appropriate balance depends on risk and intervention cost.

2.2 Economic metrics

A simple payback period is

\[

\text{Payback period}=\frac{\text{Initial investment}}{\text{Annual net cash benefit}},

\]

provided the annual benefit is reasonably stable and the numerator and denominator use consistent monetary units. This measure ignores the time value of money, asset life and benefits after payback. For major programmes, discounted cash flow, total cost of ownership and sensitivity analysis are more appropriate.

Total cost of ownership includes devices, installation, connectivity, platform fees, integration, cybersecurity, calibration, training, maintenance, replacement and decommissioning. A low-cost sensor can be uneconomic if its calibration, battery replacement or integration costs are high.

2.3 Environmental and social metrics

Relevant outcomes include energy intensity, carbon emissions, water consumption, waste, exposure reduction, accessibility, equity of service, patient outcomes, worker safety and public acceptance. Metrics must be defined operationally. For example, “energy savings” should specify the baseline, weather adjustment, production level, measurement interval and treatment of rebound effects.

A technically successful deployment may still fail socially if it excludes communities with poor connectivity, increases surveillance without safeguards, or transfers risk to workers. Evaluation must therefore combine engineering evidence with stakeholder and institutional analysis.

3. Smart Manufacturing and Predictive Maintenance

Smart manufacturing integrates machines, production systems, enterprise applications and human expertise. Typical data sources include vibration, acoustic emission, temperature, pressure, current, cycle time, quality inspection and machine-state signals. Industrial protocols may include OPC UA, MQTT, Modbus, PROFINET and time-sensitive networking, although protocol choice must reflect security and real-time requirements.

3.1 Applications

Applications include condition monitoring, predictive maintenance, adaptive process control, machine vision, production traceability, energy optimization, digital work instructions and autonomous material handling. A digital twin can combine design information, live telemetry, maintenance history and simulation to estimate remaining useful life or test process changes.

Predictive maintenance should not be confused with preventive maintenance. Preventive maintenance replaces or services components at scheduled intervals. Predictive maintenance uses condition evidence to estimate deterioration and schedule intervention when risk and cost justify it. It is most valuable when failures have significant consequences, condition indicators exist, and intervention can be planned.

A practical workflow is:

1. Define failure modes and their operational consequences.

2. Identify observable precursors and install appropriately specified sensors.

3. Establish data quality, time synchronization and asset identity.

4. Label failures and maintenance events carefully.

5. Develop a statistical or machine-learning model.

6. Validate it on future or previously unseen operating conditions.

7. Connect predictions to work orders and verify outcomes.

A model with high laboratory accuracy may fail in production because operating regimes change, failure labels are incomplete, or technicians do not trust the recommendation. Human-in-the-loop maintenance is often more robust than automatic replacement decisions.

3.2 Manufacturing constraints

Legacy equipment may lack interfaces or have proprietary data formats. Retrofitting sensors can introduce electromagnetic, mechanical or safety problems. Industrial networks require segmentation because connecting an old controller directly to an enterprise network can create a path for cyberattack. Data ownership between equipment suppliers and manufacturers may also be disputed.

Production optimization must not compromise functional safety. Safety instrumented systems and emergency controls should have independent assurance, appropriate fail-safe behaviour and carefully controlled interfaces to analytics platforms.

4. Energy and Utilities

IoT is central to smart grids, distributed energy resources, water utilities and demand-side management. Smart meters provide fine-grained consumption data; phasor measurement units observe grid dynamics; sensors monitor transformers and substations; and connected inverters coordinate solar generation, storage and electric vehicles.

A utility architecture often includes field devices, a communications network, an operations control centre, an advanced metering infrastructure, a data platform and customer or aggregator interfaces. Edge control is essential for local protection and fast balancing, while cloud analytics supports forecasting and asset management.

Applications include:

• transformer health monitoring and fault detection;
• outage localization and restoration;
• renewable generation forecasting;
• demand response and time-of-use optimization;
• leakage detection in water distribution;
• microgrid coordination; and
• electric-vehicle charging management.

Energy applications illustrate the tension between efficiency and privacy. High-resolution consumption patterns may reveal occupancy, routines and appliance use. Strong encryption, access control, aggregation, retention limits and transparent consent are required. Utilities must also consider energy justice: dynamic pricing or automated demand response should not disproportionately harm households unable to shift consumption.

Outcome metrics include technical losses, outage duration, peak demand, renewable curtailment, voltage quality, water losses, carbon intensity, customer cost and reliability indices. Metrics must distinguish improvements caused by IoT from those caused by weather, tariff changes or infrastructure investment.

5. Transportation and Logistics

Connected transportation includes intelligent traffic systems, public transport, fleet management, connected vehicles, railways, ports and warehouses. Sensors may include GPS, inertial units, cameras, radar, lidar, axle counters, weight sensors and environmental probes.

At the network level, IoT can support adaptive signal control, incident detection, parking management and travel-time prediction. At the fleet level, it supports route optimization, driver safety, fuel management, cold-chain monitoring and predictive maintenance. In ports and warehouses, asset tracking and automated handling improve throughput and visibility.

Transportation systems are safety-critical and geographically distributed. Communication loss, sensor occlusion, spoofed positioning or incorrect classification can create serious consequences. Autonomous functions therefore require defined operational design domains, fail-safe states, verification, validation and clear responsibility between manufacturers, operators and drivers.

Useful metrics include journey-time reliability, passenger waiting time, vehicle utilization, fuel or energy per tonne-kilometre, accident and near-miss rates, delivery accuracy, emissions and accessibility. Optimizing average travel time alone may increase inequity or induce additional demand. Public transport evaluation should include service coverage and reliability for vulnerable users.

6. Healthcare and Assisted Living

IoT healthcare ranges from wearable monitoring and remote patient monitoring to connected imaging, smart wards, asset tracking and medication adherence. A typical remote monitoring architecture includes a body or home sensor, a smartphone or gateway, secure communications, a clinical platform, an alert-management system and a healthcare professional.

Clinical value depends on more than measurement accuracy. A device must be clinically validated, integrated into workflow and supported by a response pathway. An alert that has no responsible recipient is not a healthcare intervention. Excessive false alarms can cause alarm fatigue, while missed events can harm patients.

Important considerations include:

• informed consent and the right to withdraw;
• protection of highly sensitive health data;
• calibration and clinical accuracy;
• cybersecurity and software updates;
• accessibility for older or disabled users;
• interoperability with electronic health records; and
• liability when an algorithm recommends or fails to recommend action.

AI may classify arrhythmias, predict deterioration or prioritize clinical worklists, but training data may underrepresent particular populations. Performance should be evaluated across relevant demographic and clinical groups. Explainability is particularly important when clinicians must justify decisions.

Outcome measures should include clinical outcomes, readmission, response time, patient-reported experience, adherence, false-alert burden and cost—not merely the number of connected devices.

7. Agriculture and Food Systems

Precision agriculture uses soil-moisture sensors, weather stations, satellite imagery, machine vision, livestock tags, irrigation controllers and farm machinery telemetry. The objective is spatially and temporally targeted management of water, nutrients, feed, pesticides and labour.

A farm IoT system must function under intermittent connectivity, harsh weather, limited power and heterogeneous equipment. Local storage and edge decisions are important. A sensor reading is useful only when its location, depth, calibration and sampling conditions are known.

Applications include variable-rate irrigation, disease detection, livestock health monitoring, greenhouse climate control, yield estimation and cold-chain traceability. Outcome metrics include yield per unit of water, fertilizer efficiency, input cost, soil health, animal welfare, post-harvest loss and profitability. Maximizing yield alone may increase chemical use, groundwater depletion or dependence on proprietary platforms.

Smallholder adoption raises questions of affordability, digital literacy, data ownership and vendor lock-in. Cooperative data platforms and open interfaces can improve bargaining power, but data-sharing arrangements must clarify who may commercialize farm data and whether it can be used for credit or insurance decisions.

8. Smart Buildings and Smart Cities

Smart buildings integrate occupancy sensing, HVAC, lighting, access control, elevators, fire systems, energy meters and building-management systems. Control strategies may reduce energy consumption by matching ventilation and temperature to occupancy. However, aggressive optimization can reduce thermal comfort or indoor-air quality.

Buildings should use separate safety and comfort control domains where appropriate. Fire detection, emergency communication and access control require high assurance and should not be compromised by experimental analytics. Building systems also have long lifecycles; procurement should require update mechanisms, documentation and interoperable interfaces.

Smart cities combine transport, utilities, waste management, public safety, environmental sensing, public spaces and civic platforms. A city is not simply a large building. It is a political and institutional system with multiple agencies, contractors and communities. Data integration may improve coordination, but centralized surveillance can undermine civil liberties.

Good city programmes begin with public problems rather than technology demonstrations. Examples include reducing bus delays, detecting water leakage, improving air-quality response or optimizing waste collection. Public dashboards, independent oversight, privacy impact assessments and community participation improve legitimacy.

9. Environmental Monitoring and Disaster Resilience

IoT supports air-quality measurement, flood and landslide warning, wildfire detection, biodiversity monitoring, coastal observation, noise mapping and industrial pollution control. Sensor networks may combine fixed stations, mobile platforms, drones, satellites and citizen measurements.

Environmental data is often affected by calibration drift, placement bias, weather, missing data and uneven spatial coverage. Low-cost sensors can provide dense observations but should be compared with reference instruments and accompanied by uncertainty information. A warning system must define thresholds, lead time, communication channels and responsible authorities. Detection without an actionable response plan has limited value.

Environmental applications raise questions of public access and accountability. Data indicating pollution or regulatory non-compliance should not be selectively withheld. At the same time, sensitive locations of endangered species may require controlled access.

10. Detailed Applied Case: Illustrative Smart Manufacturing Deployment

Background

Consider a fictional medium-sized food-processing plant with six refrigeration compressors and a packaging line. Unplanned compressor failures cause product loss, overtime and delivery delays. Maintenance is largely calendar-based, while vibration and electrical-current data are available only intermittently.

Problem

The plant proposes an IoT predictive-maintenance system. The risk is that management may purchase sensors without improving maintenance decisions. The evaluation must therefore consider failure economics, data quality, cybersecurity and workforce adoption.

Analysis

Historical records show 12 compressor-related stoppages over three years. Each stoppage causes an average direct loss of 6,000 monetary units and 10 hours of production disruption. The plant estimates that a targeted intervention costs 900 monetary units, whereas replacing components unnecessarily costs 1,500 units. These estimates are not sufficient by themselves: the analysis must also include sensor installation, platform subscriptions, calibration and integration.

Engineers identify vibration amplitude, discharge temperature, motor current and suction pressure as potential indicators. A gateway is placed in the OT network, with one-way or tightly controlled data transfer to an analytics environment. Emergency shutdown remains in the local control system.

Approach and solution

The project begins with a three-month baseline and a failure-mode analysis. Sensors are calibrated, timestamps synchronized and maintenance records standardized. A rule-based model is initially used because technicians can interpret it: combinations of rising vibration and abnormal temperature generate an inspection recommendation. Later, a supervised model is tested using time-based validation rather than random splitting, since random splitting could place data from the same failure episode in both training and test sets.

The alert policy prioritizes recall for severe failure modes but requires technician confirmation before intervention. Alerts are integrated with the computerized maintenance-management system. The plant establishes network segmentation, device identities, signed firmware updates, role-based access and a vulnerability-response process.

Outcome evaluation

After twelve months, the plant compares the intervention period with the baseline, adjusting for production volume and seasonal demand. It reports fewer emergency stoppages, lower product loss and a moderate increase in planned maintenance. The project is considered successful only if the reduction in unplanned downtime exceeds the recurring system cost and if safety, quality and worker workload remain acceptable.

Suppose the system prevents four stoppages in a year. The avoided direct loss is 4 × 6,000 = 24,000 monetary units. If six inspections are triggered and each costs 900 units, intervention cost is 5,400 units. If annual platform, calibration and support costs are 10,000 units, the estimated net direct benefit is 24,000 − 5,400 − 10,000 = 8,600 monetary units, before considering installation cost and indirect benefits. This is an illustrative calculation, not proof of causality. A controlled comparison, longer observation and sensitivity analysis are still required.

Learning points

The case demonstrates that predictive maintenance is a socio-technical intervention. Its success depends on failure-mode understanding, trustworthy data, appropriate alert thresholds, maintenance workflow, cybersecurity and economic evidence. A more complex AI model would not automatically produce greater value.

11. Cross-Cutting Constraints and Governance

Cybersecurity and resilience

IoT expands the attack surface through sensors, gateways, cloud interfaces, mobile applications and supply chains. Good practice includes asset inventories, secure identity, least privilege, network segmentation, encryption in transit and at rest, secure boot where feasible, signed updates, vulnerability management, logging, backup and incident response. Security should be designed across the lifecycle rather than added after deployment.

Resilience also requires graceful degradation. Systems should define what happens during loss of connectivity, corrupted data, clock failure, cloud unavailability or compromised devices. Manual operation and local control may be essential fallback modes.

Interoperability and lifecycle management

Standards can reduce integration costs, but “standard-compliant” does not always mean semantically interoperable. Data models, units, timestamps, quality flags and identifiers must be specified. Open APIs and export rights reduce vendor lock-in. Procurement should address end-of-support dates, replacement parts, software updates, audit rights and data portability.

Privacy, ethics and accountability

Location, health, energy and workplace data can identify individuals or infer sensitive behaviour. Privacy-by-design includes data minimization, purpose limitation, retention controls, aggregation, anonymization where appropriate and transparent user communication. Anonymization is not automatically irreversible, particularly when datasets can be linked.

Automated decisions should have defined accountability, appeal mechanisms and human oversight proportionate to risk. Workplace monitoring requires consultation and safeguards against excessive surveillance. Algorithmic systems should be assessed for bias, drift and unequal error rates.

Sustainability

The environmental impact of IoT includes device manufacture, batteries, communications, data-centre energy, replacement and electronic waste. A deployment should justify sensing density, use durable and repairable hardware, plan battery logistics and specify end-of-life recovery. More data is not necessarily better data.

12. Contemporary Developments and Best Practices

Edge AI is increasingly used where latency, privacy or connectivity makes cloud processing unsuitable. Federated learning can reduce centralization of raw data, although it introduces challenges involving communication cost, model poisoning and heterogeneous data. Digital twins are becoming more useful when linked to verified asset models and maintenance histories rather than treated as attractive visualizations.

5G and private wireless networks may support industrial mobility and low-latency services, but their value depends on application requirements and total cost. TinyML enables inference on constrained devices, reducing bandwidth and exposure of raw data. These technologies should be selected through requirements analysis, not technological fashion.

Best practices for application-oriented IoT include:

1. Start with a measurable operational or social problem.

2. Establish a baseline and define success before procurement.

3. Design the complete decision and response workflow.

4. Use the minimum sensing and retention necessary.

5. Separate safety-critical control from experimental analytics.

6. Validate models under realistic future conditions.

7. Include operators, clinicians, citizens or farmers in design.

8. Plan cybersecurity, updates, calibration and retirement from the beginning.

9. Evaluate distributional effects, accessibility and environmental cost.

10. Scale only after a pilot demonstrates reliable outcomes and a sustainable operating model.

Chapter Summary

Industrial and societal IoT connects physical assets, people and institutions to support observation, analysis and action. Its principal applications include predictive maintenance and smart manufacturing; smart grids and utilities; connected transportation; remote healthcare; precision agriculture; smart buildings and cities; and environmental monitoring. Each domain requires an architecture that balances device capability, connectivity, edge processing, cloud analytics, human decisions and actuation.

Critical evaluation depends on baselines and outcome metrics. Availability, latency, accuracy, precision, recall, downtime, cost, energy, emissions, safety, equity and user experience may all be relevant. IoT value is not established by device counts or dashboard sophistication. It is established through measurable, attributable improvement.

The major constraints are cybersecurity, interoperability, data quality, lifecycle cost, privacy, safety, regulation, workforce adoption and sustainability. Edge computing, AI, digital twins and advanced networks can strengthen applications, but only when matched to genuine requirements. Successful IoT is therefore a socio-technical and governance achievement as much as a connectivity achievement.

Key Terms

Cyber-physical system: A system in which computational processes monitor or influence physical processes.

Digital twin: A data-linked digital representation of a physical asset, process or system.

Edge computing: Processing performed near the data source rather than exclusively in a remote cloud.

Interoperability: The ability of systems to exchange and meaningfully use information.

Operational technology: Hardware and software that monitors or controls physical processes.

Predictive maintenance: Maintenance planned using evidence about asset condition or likely failure.

Preventive maintenance: Maintenance performed according to a predetermined schedule or usage interval.

Telemetry: Remote measurement and transmission of data.

Actuator: A device that produces a physical action in response to a control signal.

Digital thread: A connected flow of data linking information across an asset or product lifecycle.

Mean time between failures: Average operating time between failures of a repairable system.

Mean time to repair: Average time required to restore a failed system to service.

Precision: The proportion of positive predictions that are true positives.

Recall: The proportion of actual positive cases correctly identified.

Total cost of ownership: The complete lifecycle cost of acquiring, operating, maintaining and retiring a system.

Demand response: Adjustment of electricity consumption in response to grid conditions or prices.

Privacy by design: Integrating privacy protection into system architecture and lifecycle processes.

Digital divide: Unequal access to connectivity, devices, skills or benefits of digital services.

Graceful degradation: Continued safe operation at reduced capability during faults or disruption.

Functional safety: Reduction of risk arising from hazardous system behaviour through engineered safety functions.

Review Questions

Short Answer Questions

1. Why is a baseline necessary when evaluating an IoT deployment?

2. Distinguish preventive maintenance from predictive maintenance.

3. What is the role of edge computing in safety-critical applications?

4. Define precision and recall in the context of predictive maintenance.

5. Identify two privacy risks associated with smart-meter data.

Long Answer Questions

1. Explain a layered IoT architecture for smart manufacturing and discuss the responsibilities of each layer.

2. Critically compare IoT applications in healthcare and agriculture with respect to data quality, connectivity and human consequences.

3. Discuss how cybersecurity, interoperability and lifecycle management influence IoT procurement decisions.

4. Explain why smart-city programmes should begin with public problems rather than technology demonstrations.

5. Evaluate the economic, environmental and social metrics that should be used to assess a smart-grid project.

Analytical / Application Questions

1. A factory has frequent motor failures but incomplete maintenance records. Design a staged IoT predictive-maintenance evaluation, including data, model validation, workflow and outcome metrics.

2. A city proposes installing cameras and air-quality sensors throughout public spaces. Analyse the proposal using privacy, equity, cybersecurity, environmental and public-value criteria.

3. A remote-health system has high detection recall but generates many false alarms. Recommend a revised evaluation and governance strategy that balances patient safety, clinical workload and user trust.

Multiple Choice Questions with Answers

Question 1

Which architectural function is most appropriate for an emergency industrial shutdown?

A. Local safety and control logic

B. A public social-media platform

C. An annual cloud report

D. A remote marketing database

Correct Answer: A. Local safety and control logic

Question 2

In predictive maintenance, recall is calculated as:

A. TP divided by TP plus FP

B. TP divided by TP plus FN

C. TN divided by TN plus FP

D. FP divided by TP plus FP

Correct Answer: B. TP divided by TP plus FN

Question 3

A system has an MTBF of 900 hours and an MTTR of 100 hours. Using A = MTBF/(MTBF + MTTR), its availability is:

A. 0.10

B. 0.50

C. 0.90

D. 1.11

Correct Answer: C. 0.90

Question 4

Which outcome is most directly relevant to evaluating a demand-response programme?

A. Number of dashboard colours

B. Peak electricity demand reduction

C. Number of passwords created

D. Camera image resolution

Correct Answer: B. Peak electricity demand reduction

Question 5

Why should a healthcare IoT alert have an explicitly defined response pathway?

A. Connectivity becomes unnecessary

B. A measurement alone does not ensure clinical action

C. It eliminates the need for calibration

D. It guarantees zero false alarms

Correct Answer: B. A measurement alone does not ensure clinical action

Question 6

Which practice best reduces vendor lock-in in a city IoT programme?

A. Exclusive use of undocumented interfaces

B. Avoiding data export

C. Open APIs and defined data portability requirements

D. Prohibiting system audits

Correct Answer: C. Open APIs and defined data portability requirements

Question 7

A predictive-maintenance system prevents four failures, each costing 6,000 monetary units. Six inspections cost 900 units each, and annual platform costs are 10,000 units. What is the illustrative net direct benefit?

A. 5,400 units

B. 8,600 units

C. 14,000 units

D. 24,000 units

Correct Answer: B. 8,600 units

Question 8

Which is a major limitation of low-cost environmental sensors?

A. They can never transmit data

B. They may require comparison with calibrated reference instruments

C. They always have excessive accuracy

D. They cannot be deployed outdoors

Correct Answer: B. They may require comparison with calibrated reference instruments

Question 9

Why is random splitting of time-series failure data potentially misleading?

A. It prevents any model training

B. It may place records from the same failure episode in training and test sets

C. It always lowers sensor accuracy

D. It removes all missing values

Correct Answer: B. It may place records from the same failure episode in training and test sets

Question 10

Which principle is most consistent with privacy by design?

A. Collect and retain all data indefinitely

B. Minimize data collection and limit use to defined purposes

C. Publish identifiable health data by default

D. Delay privacy decisions until after deployment

Correct Answer: B. Minimize data collection and limit use to defined purposes