Foundations of AI, IoT and Smart Mechanical Systems

Introduction and Context

Mechanical engineering has traditionally focused on the design, manufacture, operation and maintenance of physical systems. Machines were specified through geometry, materials, mechanics, thermodynamics, fluid mechanics and control theory. Increasingly, however, mechanical systems are also expected to sense their environment, communicate information, make decisions, adapt to changing conditions and cooperate with people and other machines.

A modern industrial robot, wind turbine, automobile, machine tool or heating-and-cooling plant is therefore more than a collection of mechanical components. It is a smart mechanical system: a physical system enhanced by sensors, computation, communication, software and data-driven decision-making.

Artificial intelligence (AI), machine learning (ML), the Internet of Things (IoT), cyber-physical systems (CPS), embedded intelligence and Industry 4.0 provide the conceptual and technological foundations for this transition. These ideas are related, but they are not interchangeable. IoT primarily concerns connected sensing and communication. AI concerns systems that perform tasks associated with intelligent behaviour. A cyber-physical system integrates computation with physical processes through sensing and actuation. Embedded intelligence places computational capability close to the machine or device.

For mechanical engineers, the importance of these technologies is practical. They support predictive maintenance, energy optimization, automated inspection, digital twins, adaptive manufacturing, autonomous vehicles and human-machine collaboration. At the same time, they introduce new engineering responsibilities involving data quality, cybersecurity, safety, explainability, privacy and lifecycle management.

Learning Objectives

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

1. Define AI, machine learning, IoT, cyber-physical systems and embedded intelligence.
2. Explain how these technologies relate to mechanical engineering systems.
3. Identify the main components and information flows of a smart mechanical system.
4. Distinguish Industry 4.0 from Industry 5.0.
5. Describe representative industrial and engineering applications.
6. Evaluate benefits, limitations, risks and governance requirements.
7. Analyze a practical condition-monitoring application using an engineering systems perspective.

1. Artificial Intelligence and Machine Learning

1.1 Artificial intelligence

Artificial intelligence is the field concerned with constructing systems that perform tasks requiring capabilities commonly associated with human intelligence. Such capabilities include perception, reasoning, learning, language processing, planning and decision-making.

AI is not a single technology. It includes several approaches:

• Rule-based or symbolic AI, in which knowledge is represented through explicit rules and logical relationships.
• Search and optimization, in which a system explores possible solutions and selects one according to an objective.
• Machine learning, in which a model learns patterns from data rather than being programmed entirely through fixed rules.
• Computer vision, which enables interpretation of images or video.
• Natural-language processing, which enables analysis or generation of human language.
• Knowledge-based systems, which use structured domain knowledge to support diagnosis or decision-making.

In mechanical engineering, AI may classify surface defects, estimate remaining useful life, optimize a production schedule or recommend operating settings. AI does not eliminate engineering principles. Rather, it complements physical models, experiments and professional judgment.

1.2 Machine learning

Machine learning is a subset of AI in which an algorithm estimates a relationship from examples. A typical supervised-learning problem uses input variables, often called features, and known target values. The model is trained to minimize a loss function.

For a regression model, a general empirical-risk objective can be written as:

Here, x_i is the feature vector for observation i, y_i is its target value, f(x_i; θ) is the model prediction, θ is the set of model parameters, L is the prediction-loss function, n is the number of training observations, Ω(θ) is a regularization term, and λ is a non-negative regularization coefficient. The first term measures fit to the data; the second discourages unnecessarily complex models. The equation applies to many supervised-learning settings, but the exact loss and regularization depend on the task.

Common learning categories are:

• Supervised learning: the training data contain target outputs. Examples include predicting bearing temperature or classifying a defective component.
• Unsupervised learning: the system identifies structure without labelled targets. Examples include clustering machines according to operating behaviour and detecting unusual patterns.
• Semi-supervised learning: a small labelled dataset is combined with a larger unlabelled dataset.
• Reinforcement learning: an agent learns actions through interaction with an environment and feedback in the form of rewards or penalties. It is potentially useful for process control, but safety constraints are essential.

A model that performs well on training data may perform poorly on new data. This is overfitting. A model that is too simple to represent important relationships is affected by underfitting. Therefore, datasets should normally be separated into training, validation and test sets. Evaluation must reflect the actual operating conditions, including changes in load, temperature, speed, material and environment.

1.3 AI and physics-based engineering models

Mechanical engineering frequently possesses valuable physical knowledge. A shaft model, heat-transfer equation or vibration model may explain behaviour even when only limited data are available. A purely data-driven model can discover useful correlations, but it may fail outside the range represented in its training data.

Three broad strategies are common:

1. Physics-based modelling: use governing laws and identified parameters.
2. Data-driven modelling: learn relationships from measured data.
3. Hybrid or physics-informed modelling: combine physical constraints with learned components.

Hybrid approaches can improve interpretability and reduce the amount of data required. For example, a digital model may calculate expected bearing temperature from load and speed, while machine learning estimates the discrepancy caused by lubrication degradation. The result should still be validated experimentally and against safety requirements.

2. The Internet of Things in Mechanical Engineering

The Internet of Things is a network of physical objects equipped with sensing, processing, communication and, often, actuation capabilities. An IoT device is not merely connected hardware. It participates in a system that collects data, transfers information and supports an operational purpose.

A mechanical IoT application commonly contains the following elements:

1. Physical asset: a pump, motor, robot, compressor, vehicle or production line.
2. Sensors: devices that measure variables such as temperature, pressure, force, displacement, vibration, current or flow.
3. Signal-conditioning electronics: circuits that amplify, filter, isolate or digitize signals.
4. Embedded processor: hardware that performs local computation and manages the device.
5. Communication network: wired or wireless infrastructure that transports data.
6. Data platform: local, edge or cloud storage and processing resources.
7. Analytics and decision layer: dashboards, models, alarms or optimization software.
8. Actuators and control system: devices that change speed, position, flow, torque or other physical variables.
9. Human and organizational interface: operators, maintenance personnel, engineers and management systems.

IoT architectures are frequently described in layers. A practical five-layer view is:

The distinction between edge computing and cloud computing is important. Edge computing processes data close to the machine. It can reduce latency, bandwidth use and dependence on a remote connection. Cloud computing offers scalable storage and computational resources and facilitates fleet-wide analysis. Many industrial systems use both: rapid protective actions at the edge and broader optimization in a central platform.

2.1 Measurement quality

An IoT system is only as useful as the quality and relevance of its measurements. Important sensor characteristics include:

• Accuracy: closeness of a measured value to the accepted reference value.
• Precision: repeatability of measurements.
• Resolution: smallest detectable change.
• Range: interval over which the sensor operates acceptably.
• Sampling rate: number of measurements acquired per unit time.
• Drift: gradual change in sensor response over time.
• Latency: delay between the physical event and the available measurement.

Sampling must be appropriate for the physical phenomenon. If a vibration signal contains important high-frequency components, a slow sampling rate can hide or distort them. Sensor placement is equally important: a temperature sensor mounted far from a bearing may measure housing temperature rather than the bearing condition of interest.

3. Cyber-Physical Systems and Embedded Intelligence

A cyber-physical system integrates physical processes with computation, communication and feedback. The term emphasizes the interaction between the cyber domain—software, data and networks—and the physical domain—materials, energy, motion and forces.

A conventional mechanical system may operate according to fixed settings. A CPS observes its state, compares the observed state with objectives or constraints, computes a response and applies that response through actuators. The loop is therefore:

physical process → sensing → computation → communication and decision → actuation → physical process

The loop may operate at several time scales. A motor drive may execute a protective control loop in microseconds or milliseconds, while a maintenance-planning system may analyze trends over weeks. These layers must be coordinated. A cloud-based prediction that arrives after a machine has failed cannot replace a local safety interlock.

3.1 Embedded intelligence

Embedded intelligence is computational capability integrated into a product or machine rather than supplied solely by a separate computer. A smart actuator may measure its own position, compensate for nonlinearities and report health status. A smart bearing housing may contain vibration sensing and local anomaly detection.

Embedded systems are constrained by processor capability, memory, electrical power, thermal conditions, communication bandwidth and cost. They may need to operate in real time and in locations where network connectivity is intermittent. Consequently, an algorithm suitable for a data centre may need to be simplified, compressed or redesigned for an embedded controller.

Important design requirements include:

• deterministic timing for safety-critical control;
• secure boot and authenticated software updates;
• fault detection and safe fallback behaviour;
• electromagnetic compatibility;
• resistance to temperature, shock, dust and moisture;
• maintainability over the machine’s service life.

3.2 Control and AI: complementary roles

Classical control remains fundamental. A feedback controller measures an output, compares it with a reference and adjusts an input. For a simple single-input, single-output system, the error is:

where r(t) is the desired reference, y(t) is the measured output and e(t) is the control error. The quantities must have compatible units.

A proportional-integral-derivative controller may be represented by:

where u(t) is the control input, K_p, K_i and K_d are controller gains, and τ is an integration variable. The derivative term is sensitive to measurement noise, and practical implementations commonly filter it. This equation describes a continuous-time ideal PID controller; actual digital implementations use sampling and numerical approximations.

AI can assist control by estimating unmeasured states, tuning parameters, detecting faults or optimizing operating points. It should not be inserted into a safety-critical loop without analysis of stability, failure modes, verification and fallback operation.

4. Industry 4.0 and Industry 5.0

4.1 Industry 4.0

Industry 4.0 describes the fourth industrial transformation, characterized by connected equipment, automation, cyber-physical systems, data analytics, digital platforms and increasingly autonomous decision-making. Its objective is not simply to install sensors; it is to create integrated and responsive production systems.

Typical Industry 4.0 capabilities include:

• machine-to-machine communication;
• interoperable production systems;
• real-time production visibility;
• digital twins and simulation;
• predictive maintenance;
• flexible and mass-customized production;
• autonomous material handling;
• integrated enterprise and supply-chain data.

A digital twin is a digital representation of a physical asset or process that is linked to the physical system through data. A static computer-aided design model is not necessarily a digital twin. The representation must have an appropriate connection to the asset and must support a defined purpose, such as monitoring, prediction, optimization or training.

4.2 Industry 5.0

Industry 5.0 extends the technology-centred focus of Industry 4.0 by emphasizing human-centricity, sustainability and resilience. It asks not only whether a factory can be automated, but also whether technology improves human work, reduces environmental impact and enables robust operation under disruption.

Industry 5.0 does not replace Industry 4.0. Connectivity, automation and data analytics remain important, but they are evaluated through broader social and environmental objectives. Examples include collaborative robots that assist rather than displace skilled workers, production planning that accounts for energy and material use, and systems designed to continue operating safely when supply chains or networks are disrupted.

5. A Reference Architecture for a Smart Mechanical System

A useful engineering analysis separates the system into functional blocks.

Physical layer

This includes structures, mechanisms, motors, transmissions, fluids, thermal components and workpieces. Mechanical performance remains governed by established principles such as force balance, energy conservation, material behaviour, friction, wear and heat transfer.

Perception layer

Sensors convert physical quantities into electrical or digital signals. Sensor selection must follow the engineering question. For example, detecting imbalance may require vibration measurements, whereas detecting overheating may require temperature and electrical-current measurements.

Computation layer

Local controllers, gateways or servers filter signals, calculate features, execute models and manage data. Typical vibration features include root-mean-square amplitude, peak value, crest factor and frequency-domain amplitudes. A feature is useful only if it is related to a failure mechanism or decision requirement.

Communication layer

Communication may use industrial Ethernet, fieldbus, Wi-Fi, cellular networks or low-power wireless systems. Selection depends on latency, reliability, distance, bandwidth, environmental conditions, security and cost.

Intelligence layer

This layer performs classification, prediction, diagnosis, optimization or planning. It may use rules, statistical models, machine learning or a hybrid method.

Action and governance layer

The output may be an alarm, maintenance work order, changed machine setting or automatic shutdown. Governance determines who can authorize actions, how decisions are recorded and how performance is audited.

A complete system must also include lifecycle activities: calibration, data management, software updates, cybersecurity monitoring, spare-parts planning, decommissioning and disposal.

6. Applications in Mechanical Engineering

6.1 Predictive maintenance

Predictive maintenance uses condition information to estimate whether maintenance is needed and when it should occur. It differs from preventive maintenance, which schedules maintenance according to time or usage, and corrective maintenance, which acts after failure.

A typical workflow is:

1. Define the failure modes and consequences.
2. Select measurable indicators linked to those failure modes.
3. Acquire and validate data.
4. Establish healthy operating baselines.
5. Detect anomalies or estimate degradation.
6. Confirm the diagnosis through engineering inspection.
7. Schedule an intervention and record the outcome.

The aim is not to maximize the number of alarms. It is to support safe, economical and timely decisions.

6.2 Smart manufacturing

Connected machine tools can report spindle load, vibration, tool wear and dimensional quality. AI can estimate tool life, while process models can relate cutting speed, feed and depth of cut to productivity and surface integrity. Automated inspection using machine vision can identify scratches, burrs, missing components or dimensional deviations.

6.3 Robotics and autonomous systems

Robots combine mechanical structures, actuators, sensors, control algorithms and planning software. AI supports object recognition, motion planning, grasp selection and anomaly detection. In collaborative robotics, the design must also address force limitation, workspace monitoring, predictable behaviour and human-machine communication.

6.4 Energy and thermal systems

Smart HVAC systems can use occupancy, weather and equipment data to reduce energy use while maintaining thermal comfort. Industrial energy systems can detect compressed-air leakage, optimize chiller operation and identify inefficient pumps. Any optimization must respect constraints such as pressure, temperature, safety limits and equipment ratings.

6.5 Transportation and mobile machinery

Vehicles and construction equipment use sensors and embedded processors for engine management, driver assistance, fleet monitoring and route optimization. Mechanical engineers contribute knowledge of dynamics, loads, fatigue, thermal management, braking and failure modes.

7. Benefits and Limitations

Potential benefits include:

• reduced unplanned downtime;
• improved product quality and traceability;
• more efficient energy and material use;
• faster fault diagnosis;
• improved worker safety;
• flexible and customized production;
• better lifecycle planning;
• new service models based on equipment performance.

However, these benefits are not automatic. Important limitations include:

• insufficient or biased data;
• sensor failure and calibration drift;
• false alarms and missed detections;
• changing operating conditions, called concept drift;
• dependence on network and electrical infrastructure;
• cybersecurity vulnerabilities;
• integration problems between old and new equipment;
• unclear return on investment;
• lack of skilled personnel;
• poor interpretability of complex models;
• unsafe actions caused by incorrect predictions.

An AI model can identify correlation without establishing causation. For example, high vibration may be associated with bearing damage, but it may also result from imbalance, looseness, resonance or a changed foundation. Diagnosis should therefore combine data analysis with physical reasoning and inspection.

8. Governance, Safety and Ethics

Smart mechanical systems operate within social, legal and professional contexts. Safety is the first requirement where failure can injure people, damage equipment or release hazardous energy. AI-generated recommendations should be subject to defined authority, verification and safe-state procedures.

Cybersecurity is an engineering concern because a malicious command can alter speed, pressure or position. Good practice includes network segmentation, least-privilege access, encryption where appropriate, strong authentication, secure updates, vulnerability management, logging and incident response.

Data governance should specify ownership, retention, access, quality standards and permitted uses. Industrial data may reveal production rates, proprietary processes or worker activities. Systems should avoid unnecessary surveillance and should communicate how data are used.

Important AI governance principles include:

• Validity and reliability: performance must be demonstrated in the intended operating domain.
• Transparency: users should understand the model’s purpose, inputs and limitations.
• Accountability: responsibility for decisions must remain identifiable.
• Human oversight: people must be able to intervene where consequences are significant.
• Fairness: systems affecting workers or access to opportunities should be checked for unjustified bias.
• Security and privacy: data and models must be protected.

Standards and regulations vary by jurisdiction and application. Engineers must consult applicable machinery-safety, functional-safety, cybersecurity, data-protection and quality-management requirements rather than assuming that a technically accurate model is automatically legally acceptable.

9. Illustrative Case: Predictive Monitoring of an Industrial Pump

Background

Consider a centrifugal pump used to circulate cooling water in a manufacturing plant. Unexpected failure stops production and may cause overheating of downstream equipment. The pump has an electric motor, shaft, coupling, bearings, impeller and seals.

Problem

The maintenance department currently replaces bearings every six months. Some bearings are replaced while still healthy, while other failures occur between inspections. The plant wants earlier warning without creating excessive false alarms.

Analysis

Possible failure modes include bearing wear, shaft misalignment, imbalance, cavitation, seal leakage and motor electrical faults. The engineering team maps each failure mode to observable indicators:

Three-axis vibration sensors are mounted near the drive-end and non-drive-end bearings. Temperature and motor current are also recorded. The sampling strategy is selected according to the frequency range needed for bearing and rotating-equipment diagnosis. Operating speed, flow rate and load are recorded because vibration levels depend on operating conditions.

Approach and solution

A staged approach is selected rather than immediate deployment of a complex neural network.

1. The team establishes a healthy baseline during confirmed normal operation.
2. Data are labelled using maintenance records and inspection results.
3. Signal-quality checks identify missing samples, sensor saturation and impossible values.
4. Baseline statistical limits provide a transparent first-level alarm.
5. A supervised classifier is trained only after sufficient labelled examples are available.
6. The model output is combined with operating context and reviewed by a reliability engineer.
7. A high-confidence warning creates an inspection request, not an automatic shutdown.
8. A separate hardwired protection system remains responsible for immediate unsafe conditions.

The system is evaluated using missed failures, false alarms, warning time and maintenance cost. Accuracy alone is insufficient. If a rare failure occurs only once, a model can appear accurate by predicting “healthy” most of the time. The test set must represent realistic future operating conditions and must be separated chronologically where appropriate.

Outcome

In this illustrative case, the monitoring system detects an increasing vibration trend and recommends inspection. The engineer finds early misalignment rather than bearing damage. The coupling is corrected during planned downtime, avoiding a secondary bearing failure. The result demonstrates that AI need not produce a final diagnosis independently; its value may be to prioritize attention and improve the timing of engineering decisions.

Learning points

• Failure-mode analysis should precede sensor and model selection.
• Operating context is essential for interpreting measurements.
• A transparent baseline can be valuable before advanced AI is introduced.
• Human inspection and physical reasoning remain necessary.
• Safety protection and predictive analytics should be treated as distinct functions.
• Success should be measured through operational outcomes, not only model metrics.

10. Best Practices for Mechanical Engineers

A reliable implementation generally follows these principles:

1. Begin with a defined engineering decision, not with a fashionable technology.
2. Identify failure modes, physical constraints and safety consequences.
3. Select sensors based on observability, placement, bandwidth and environmental suitability.
4. Establish data-quality procedures, calibration schedules and metadata standards.
5. Start with interpretable baselines and compare advanced models against them.
6. Separate development, validation and operational monitoring.
7. Test under changing loads, speeds, temperatures and production conditions.
8. Design for cybersecurity, maintainability and graceful degradation.
9. Include operators and maintenance personnel in system design.
10. Quantify benefits, costs, false alarms, downtime and lifecycle effects.
11. Document assumptions, model versions, training data and decision authority.
12. Reassess performance after equipment modification, process change or sensor replacement.

11. Contemporary Developments

Current development is moving toward edge AI, foundation models for industrial data, autonomous mobile robots, federated learning, low-code industrial analytics and increasingly integrated digital twins. These developments may reduce deployment time, but they do not remove the need for domain expertise.

Industrial foundation models may process vibration, images, maintenance text and process histories together. Federated learning can allow several sites to contribute to model improvement without directly pooling all raw data, although it introduces additional concerns about security, statistical differences and governance. Digital twins are becoming more useful when simulation, live measurements and lifecycle records are connected through consistent data models.

The most durable professional capability remains the ability to connect algorithmic outputs with physical mechanisms, business objectives, safety requirements and human work. Mechanical engineers who understand both machine behaviour and information systems are well positioned to design and validate smart products and factories.

Chapter Summary

AI enables machines and software to perform tasks involving perception, learning, reasoning and decision-making. Machine learning is one important AI approach, but its performance depends on representative data, appropriate evaluation and control of overfitting. IoT connects physical assets through sensors, processors, networks, platforms and applications. Cyber-physical systems integrate these computational functions with physical processes through feedback, while embedded intelligence places processing capability close to the machine.

Industry 4.0 emphasizes connected, automated and data-integrated production. Industry 5.0 extends this vision through human-centricity, sustainability and resilience. Smart mechanical systems require more than sensors and algorithms: they require sound mechanical analysis, reliable measurement, suitable communication, safe actuation, cybersecurity, governance and lifecycle management.

Applications include predictive maintenance, smart manufacturing, robotics, energy management and intelligent transportation. The benefits can be substantial, but limitations include poor data, sensor faults, changing conditions, false alarms, cybersecurity threats, integration costs and lack of explainability. The pump case demonstrates that the most effective approach combines failure-mode analysis, condition monitoring, interpretable analytics and professional engineering judgment.

Key Terms

Artificial intelligence: The field of creating systems capable of tasks associated with intelligent human behaviour.

Machine learning: An AI approach in which models learn patterns or relationships from data.

Internet of Things: A network of physical objects equipped with sensing, processing and communication capabilities.

Cyber-physical system: An integrated system in which computation, communication and physical processes interact through feedback.

Embedded intelligence: Computational capability incorporated directly into a device or machine.

Sensor: A device that measures a physical quantity and produces a usable signal.

Actuator: A device that converts a control signal into physical action.

Edge computing: Processing data near the physical asset that generates it.

Cloud computing: On-demand computing and storage supplied through networked infrastructure.

Digital twin: A connected digital representation of a physical asset or process used for monitoring, prediction or optimization.

Industry 4.0: Connected and data-driven industrial transformation based on automation and cyber-physical integration.

Industry 5.0: An industrial vision emphasizing human-centricity, sustainability and resilience alongside advanced technology.

Predictive maintenance: Maintenance planned using condition information or predictions of degradation and failure.

Overfitting: A condition in which a model fits training data well but generalizes poorly to new data.

Concept drift: A change over time in the relationship between inputs and the target or operating behaviour.

Digital thread: A connected flow of lifecycle information across design, manufacture, operation and maintenance.

Interoperability: The ability of systems, devices or organizations to exchange and use information effectively.

Explainability: The extent to which the basis of a model’s output can be understood by relevant users.

Functional safety: The reduction of risk through safety-related control functions and specified failure responses.

Condition monitoring: The measurement and interpretation of equipment condition to support maintenance or safe operation.

Review Questions

Short Answer Questions

1. Define artificial intelligence and distinguish it from machine learning.
2. What are the main functional layers of an IoT-based mechanical system?
3. Explain the meaning of a cyber-physical system.
4. State two differences between Industry 4.0 and Industry 5.0.
5. Why is sensor quality important in predictive maintenance?

Long Answer Questions

1. Explain the relationship among AI, IoT, embedded intelligence and cyber-physical systems, using a mechanical-engineering example.
2. Discuss the roles of edge computing and cloud computing in an industrial monitoring system.
3. Compare physics-based, data-driven and hybrid modelling approaches for mechanical equipment.
4. Describe the benefits and limitations of AI-enabled predictive maintenance.
5. Explain the safety, cybersecurity, ethical and governance issues that arise in smart mechanical systems.

Analytical / Application Questions

1. A factory wants to monitor a gearbox for tooth damage, imbalance and lubrication failure. Propose a sensing, data-processing and decision architecture, and justify each major design choice.
2. A machine-learning model reports 98% accuracy for a rare equipment failure, but almost all test examples are healthy machines. Analyze why this result may be misleading and propose a better evaluation strategy.
3. Design an implementation plan for converting a conventional pump station into a smart mechanical system. Include failure-mode analysis, sensors, edge functions, communication, AI use, human oversight, cybersecurity and performance measures.

Multiple Choice Questions with Answers

Question 1

Which statement best describes a cyber-physical system?

A. A mechanical system with no sensors or software

B. A system integrating computation and communication with physical processes through feedback

C. A cloud database used only for storing maintenance documents

D. A machine that operates exclusively through manual control

Correct Answer: B. A system integrating computation and communication with physical processes through feedback

Question 2

Which activity is an example of supervised machine learning?

A. Grouping machines without any target labels

B. Learning a fault classifier from vibration data labelled as healthy or faulty

C. Selecting a random maintenance date

D. Measuring temperature with an uncalibrated sensor

Correct Answer: B. Learning a fault classifier from vibration data labelled as healthy or faulty

Question 3

What is a principal advantage of edge computing in an industrial machine?

A. It always eliminates the need for sensors

B. It can reduce latency and dependence on a remote network connection

C. It guarantees that every prediction is correct

D. It removes all cybersecurity requirements

Correct Answer: B. It can reduce latency and dependence on a remote network connection

Question 4

Which feature is most directly associated with Industry 5.0?

A. Human-centricity, sustainability and resilience

B. Complete replacement of all workers by machines

C. Removal of communication networks from factories

D. Exclusive use of handwritten production records

Correct Answer: A. Human-centricity, sustainability and resilience

Question 5

A model performs very well on training data but poorly on new operating data. What problem is most likely present?

A. Overfitting

B. Calibration

C. Actuation

D. Interoperability

Correct Answer: A. Overfitting

Question 6

Which statement about a digital twin is most accurate?

A. It is always identical to a computer-aided design drawing

B. It is a connected digital representation linked to a physical asset for a defined purpose

C. It is a replacement for all physical inspection

D. It is only a wireless temperature sensor

Correct Answer: B. It is a connected digital representation linked to a physical asset for a defined purpose

Question 7

Why should operating context such as speed and load be recorded during pump vibration monitoring?

A. Vibration behaviour can depend on operating conditions

B. Operating context makes sensors unnecessary

C. Speed and load are unrelated to mechanical behaviour

D. Context data automatically prove causation

Correct Answer: A. Vibration behaviour can depend on operating conditions

Question 8

Which action is most appropriate for a high-confidence predictive-maintenance warning in the illustrative pump case?

A. Ignore it because all AI outputs are unreliable

B. Immediately remove every pump from service without inspection

C. Create an inspection request while retaining independent safety protection

D. Disable the hardwired protection system

Correct Answer: C. Create an inspection request while retaining independent safety protection

Question 9

Which statement best distinguishes preventive from predictive maintenance?

A. Preventive maintenance is scheduled by time or usage, while predictive maintenance uses condition information

B. Preventive maintenance occurs only after failure, while predictive maintenance never uses sensors

C. The two terms always mean exactly the same thing

D. Predictive maintenance is limited to lubrication tasks

Correct Answer: A. Preventive maintenance is scheduled by time or usage, while predictive maintenance uses condition information

Question 10

What is the primary purpose of regular sensor calibration?

A. To increase network bandwidth

B. To maintain confidence that measured values correspond adequately to physical quantities

C. To replace the machine controller

D. To guarantee that a machine-learning model cannot overfit

Correct Answer: B. To maintain confidence that measured values correspond adequately to physical quantities