Contemporary Applications and Future Directions

Introduction

The Internet of Things (IoT) in mechatronics has progressed from connecting sensors and actuators to creating autonomous, cooperative, and increasingly intelligent cyber–physical systems. A modern mechatronic product may contain embedded computation, wireless communication, machine-learning models, real-time control, digital-twin services, and secure cloud or edge infrastructure. These technologies enable predictive maintenance, flexible manufacturing, autonomous logistics, collaborative robots, smart energy systems, and highly personalized products.

However, technological capability alone does not determine engineering value. A system must also be safe, explainable, maintainable, interoperable, energy-efficient, secure, and acceptable to its users. This chapter evaluates important contemporary directions in IoT-enabled mechatronics: autonomous and collaborative systems; human-centric Industry 5.0; 5G, 6G, and time-sensitive networking; federated learning and TinyML; swarm systems; distributed ledgers; energy harvesting; interoperable digital twins; and quantum-resilient security.

Learning Objectives

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

1. Distinguish autonomous, collaborative, and human-centric mechatronic systems.
2. Explain how 5G, 6G concepts, and time-sensitive networking support industrial communication.
3. Compare cloud, edge, federated, and TinyML approaches to industrial intelligence.
4. Evaluate swarm robotics, distributed ledgers, energy harvesting, and digital twins.
5. Identify cybersecurity risks and explain the need for post-quantum protection.
6. Propose an integrated architecture for a safe and interoperable IoT-enabled mechatronic application.
7. Critically assess technical, organizational, ethical, and regulatory limitations.

1. From Connected Devices to Autonomous and Collaborative Systems

A connected system transmits data. An autonomous system additionally perceives its environment, makes decisions, and acts with limited human intervention. A collaborative system is designed to coordinate with other machines, humans, or software agents. These categories overlap but are not identical.

A useful control-loop abstraction is:

Here, s_t is the system state at time t. Sensors provide incomplete and noisy observations; estimation combines measurements with a model; the decision layer selects an action; and actuation changes the physical state. The abstraction applies to mobile robots, autonomous vehicles, adaptive machines, and intelligent production cells.

Autonomy is not a binary property. It can be classified by the consequences of failure and the degree of human authority:

In mechatronics, autonomy requires more than machine learning. It depends on mechanical design, sensing, actuation, state estimation, control stability, communications, safety engineering, and fault management. An accurate classifier cannot compensate for an inadequately sized actuator or an unsafe failure mode.

Collaborative robots, or cobots, illustrate the distinction between automation and collaboration. A conventional industrial robot is often separated from people by fencing. A cobot may share a workspace, but safe collaboration requires speed and separation monitoring, force limitation, protective stops, reliable sensing, and risk assessment. The engineering objective is not simply to remove barriers; it is to control the probability and severity of harm.

A collaborative system should define:

• the human and machine tasks;
• permissible contact forces and speeds;
• authority during conflicting commands;
• communication and warning methods;
• degraded modes after sensor or network failure; and
• recovery procedures after an emergency stop.

2. Industry 5.0: Human-Centric and Sustainable Production

Industry 4.0 emphasized cyber–physical systems, connectivity, automation, and data-driven production. Industry 5.0 extends this perspective by emphasizing human centricity, sustainability, and resilience. It does not necessarily replace Industry 4.0; rather, it asks whether digital technologies improve human work and societal outcomes.

Human centricity

Human-centric production treats workers as skilled partners rather than sources of variability to be eliminated. Examples include exoskeleton assistance, adaptive interfaces, cobots that respond to human intention, and decision-support systems that preserve human authority in unusual situations. A human-centric design should account for cognitive workload, accessibility, training, privacy, and the possibility that automation changes rather than removes work.

Sustainability

IoT systems can reduce waste through predictive maintenance, energy monitoring, optimized logistics, and product-life extension. They can also increase environmental impact through additional electronics, energy consumption, data-center demand, rare-material extraction, and electronic waste. A valid sustainability assessment therefore considers the entire life cycle: manufacture, deployment, operation, maintenance, upgrade, reuse, and disposal.

Resilience

Resilience is the capacity to anticipate, withstand, recover from, and adapt to disruptions. Relevant disruptions include cyberattacks, component shortages, network outages, extreme weather, and defective software updates. Modular architectures, local fallback control, diverse suppliers, spare-part strategies, and tested recovery plans improve resilience.

A key critical issue is responsibility. If an AI-enabled machine makes an unsafe recommendation, responsibility may be distributed among the equipment manufacturer, software developer, system integrator, operator, and organization deploying it. Governance must therefore document assumptions, training data, authority boundaries, validation evidence, and change history.

3. 5G, 6G, and Time-Sensitive Networking

Industrial IoT communication has competing requirements. A mobile robot may need mobility and broad coverage, while a coordinated motion-control loop requires bounded latency, low jitter, high availability, and precise timing.

For a message, end-to-end latency can be represented as:

where the terms represent sensing, queueing, transmission, computation, and actuation delays. The equation is a decomposition, not a universal performance guarantee. Each term may vary, and the control system must be designed for worst-case or statistically bounded behavior rather than only average latency.

5G industrial features may include enhanced mobile broadband, massive machine-type communication, and ultra-reliable low-latency communication. In practice, performance depends on spectrum, radio conditions, network configuration, traffic load, device capability, and the application’s safety requirements. A label such as “5G” does not automatically guarantee deterministic control.

6G remains a research and standardization direction rather than a universally deployed technology. Proposed themes include integrated sensing and communication, stronger use of artificial intelligence, higher-frequency operation, distributed intelligence, and tighter integration of terrestrial and non-terrestrial networks. Engineers should distinguish established capabilities from research claims.

Time-sensitive networking (TSN) is a family of Ethernet mechanisms intended to provide predictable communication. Important concepts include synchronized clocks, scheduled traffic, traffic shaping, frame pre-emption, and explicit paths. TSN is especially useful where periodic control traffic must coexist with ordinary monitoring and diagnostic traffic.

The choice between industrial wireless and TSN is not simply a competition. A factory may use TSN for fixed high-criticality equipment, private 5G for mobile assets, and conventional Ethernet or Wi-Fi for noncritical data. A systems engineer should map each traffic class to its required bandwidth, maximum latency, jitter, reliability, security, and mobility.

4. Edge Intelligence, Federated Learning, and TinyML

Cloud computing provides scalable storage and powerful computation, but sending all industrial data to a remote cloud can introduce latency, bandwidth cost, privacy risk, and dependence on connectivity. Edge computing places processing near the machine. TinyML places compact machine-learning inference directly on microcontrollers or low-power embedded devices.

A practical architecture may contain four layers:

1. Device layer: sensing, actuation, signal conditioning, and local protection.
2. Edge layer: filtering, feature extraction, rapid inference, and coordination.
3. Site layer: production analytics, digital twins, and fleet management.
4. Cloud or enterprise layer: long-term learning, optimization, and reporting.

TinyML is appropriate when the task is narrow and the input can be represented compactly, such as detecting vibration anomalies, recognizing a motor’s acoustic state, or identifying a valve’s operating condition. It is constrained by memory, processing capacity, energy, model size, and training-data quality. Quantization and pruning can reduce resource requirements, but they may reduce accuracy.

Federated learning trains a shared model without centrally collecting all raw data. If client model parameters or updates are w_i and client data sizes are n_i, a common weighted aggregation is:

Here, K is the number of participating clients. This expression assumes that the updates are compatible, that the aggregation server can receive them, and that data distributions and optimization conditions are sufficiently manageable. It does not by itself guarantee privacy or robustness.

Industrial data are often non-independent and non-identically distributed. Different machines may have different loads, ages, suppliers, and operating environments. This non-IID condition can make a global model less accurate for minority machines. Additional concerns include poisoned updates, inference attacks, unreliable participants, and communication overhead. Secure aggregation, anomaly detection, differential privacy, and careful validation may be required.

Federated learning is therefore best viewed as a governance and systems problem as well as a machine-learning technique. The organization must define who owns the data, who may contribute updates, how models are validated, and how a faulty model is rolled back.

5. Swarm Systems and Distributed Coordination

Swarm systems use many relatively simple agents whose local interactions produce coordinated global behavior. Examples include warehouse robots, inspection drones, agricultural vehicles, and underwater sensor platforms. Swarm approaches can provide scalability, redundancy, and graceful degradation.

Typical swarm principles include:

• local sensing rather than complete global knowledge;
• neighbor-to-neighbor communication;
• distributed task allocation;
• separation, alignment, and cohesion;
• redundancy and reconfiguration; and
• emergent behavior from simple rules.

The central challenge is verification. A swarm can display unexpected collective behavior even when each individual robot appears safe. Communication loss, heterogeneous capabilities, congestion, and malicious agents can cause oscillation, deadlock, or concentration in hazardous areas.

Swarm systems should therefore include explicit constraints: geofencing, collision avoidance, maximum density, role limits, emergency return behavior, and human override. Simulation is valuable, but simulation-to-reality differences must be tested through staged deployment.

6. Distributed Ledgers in Industrial IoT

A distributed ledger records transactions across multiple participants, often using cryptographic techniques to make unauthorized alteration detectable. In IoT, possible uses include component provenance, maintenance records, calibration certificates, energy transactions, and machine-to-machine agreements.

A ledger is useful when several organizations need a shared record but do not fully trust one another. It is less useful when a single organization already controls a reliable database. A ledger does not automatically prove that a sensor reading is true: it can preserve a false measurement just as reliably as a true one. This is the “oracle” problem between the physical world and the digital record.

Industrial evaluation should consider:

• transaction throughput and latency;
• storage growth and data retention;
• identity and access control;
• confidentiality of commercial information;
• governance and dispute resolution;
• energy consumption of the consensus mechanism; and
• legal status of records and signatures.

Large sensor streams should rarely be placed directly on a ledger. A more practical design stores hashes, permissions, or references on the ledger while retaining bulk data in controlled repositories.

7. Energy Harvesting and Sustainable Embedded Operation

Energy harvesting converts ambient energy into electrical energy. Sources include vibration, thermal gradients, light, radio-frequency fields, and fluid motion. Harvesting is especially attractive for sensors that are difficult or expensive to access for battery replacement.

A basic energy balance is:

The quantities are energies over a defined time interval and must use consistent units, such as joules. The inequality is a feasibility condition, not a guarantee of continuous operation. Ambient energy is variable, conversion efficiency is less than 100%, and storage losses and startup requirements matter.

For a photovoltaic harvester under approximately constant conditions:

where P_e is electrical power in watts, η is dimensionless conversion efficiency, A is active area in square metres, and G is incident irradiance in watts per square metre. The relation is an estimate under the stated conditions; temperature, orientation, spectral content, and power-electronics losses can reduce actual output.

Energy-aware IoT design uses duty cycling, event-triggered sensing, local feature extraction, compressed communication, and adaptive sampling. A sensor node should not transmit high-rate raw data when a local model can identify the relevant event. Nevertheless, safety-critical measurements may require redundancy and minimum sampling rates that cannot be reduced solely for energy savings.

8. Interoperable Digital Twins

A digital twin is a computational representation connected to a physical asset through data and models over an appropriate life cycle. It may contain geometry, state variables, operating history, simulation models, maintenance information, and control interfaces.

A useful distinction is between:

• a digital model, which may not exchange data with the physical asset;
• a digital shadow, in which data flow is mainly from physical asset to digital representation; and
• a digital twin, in which data and actions may flow in both directions.

Interoperability has several dimensions. Syntactic interoperability concerns data formats and interfaces. Semantic interoperability concerns whether terms mean the same thing. Organizational interoperability concerns processes, authority, and agreements. A twin that exchanges syntactically valid but semantically inconsistent data is not truly interoperable.

Digital twins should define asset identity, units, coordinate systems, timestamps, uncertainty, provenance, update frequency, and version history. Open information models and standardized interfaces can reduce vendor lock-in, but standards do not eliminate the need for engineering mapping between different models.

A twin may support condition monitoring, virtual commissioning, what-if analysis, energy optimization, and operator training. It can also create risk if stale or inaccurate models are used for control. Critical decisions should include model-confidence indicators and safe fallback behavior.

9. Quantum-Resilient Security

Many current security systems rely on mathematical problems that are difficult for classical computers, such as integer factorization or discrete logarithms. Large-scale quantum computers could threaten some public-key methods through algorithms such as Shor’s algorithm. Symmetric cryptography is affected differently; larger key sizes can generally increase resistance to exhaustive search, although complete security analysis remains necessary.

Quantum-resilient, or post-quantum, security uses algorithms designed to resist known classical and quantum attacks. Candidate families include lattice-based, hash-based, code-based, and other constructions. Engineering choices must consider key and signature size, computational cost, memory, bandwidth, implementation maturity, and certification requirements.

The threat is relevant before a quantum computer exists because adversaries may collect encrypted traffic now and attempt decryption later. This is especially important for industrial assets with long service lives. A migration strategy should include:

1. inventory of cryptographic algorithms and certificates;
2. identification of long-lived confidential data;
3. use of cryptographic agility so algorithms can be replaced;
4. hybrid transition mechanisms where appropriate;
5. secure firmware and certificate-update procedures; and
6. testing on constrained devices and real-time networks.

Security is broader than encryption. Device identity, secure boot, signed updates, least privilege, network segmentation, logging, vulnerability management, and incident response remain essential.

10. Integrated Illustrative Case: Smart Collaborative Assembly Cell

This is a fictional illustrative case. A manufacturer assembles customized electric-drive modules. It intends to introduce a mobile robot, a cobot, edge-based quality inspection, a digital twin, and predictive maintenance while preserving worker authority.

Background and problem

The existing line uses fixed automation and manual inspection. Product variants cause frequent changeovers, unexpected motor vibration creates downtime, and maintenance records are distributed across incompatible systems. Management wants flexibility but cannot accept uncontrolled robot motion or the transfer of proprietary production data to an external cloud.

Analysis

The application contains several traffic and control classes. Safety functions require independently validated mechanisms and should not depend exclusively on a remote service. Motion coordination requires bounded timing. Vibration analysis can tolerate local processing and occasional model updates. Maintenance data require traceability but not necessarily a blockchain.

Approach and solution

The proposed architecture uses:

• local safety-rated sensing and control for emergency and protective functions;
• TSN Ethernet for fixed, time-sensitive equipment;
• private industrial wireless for the mobile robot;
• an edge computer for vision inference and vibration analysis;
• TinyML on selected sensors for low-power anomaly detection;
• federated learning across production cells, with model validation before deployment;
• a digital twin using common asset identifiers, units, timestamps, and versioned interfaces;
• a conventional governed database for high-volume data; and
• a permissioned ledger only for selected maintenance and calibration attestations.

The cobot operates under defined speed, separation, and force limits. Operators can pause the task, change production priorities, and request manual inspection. The robot enters a safe degraded mode if localization confidence, communication quality, or sensor health falls below an approved threshold.

Outcome

The integrated design reduces unnecessary raw-data transmission and allows rapid local decisions. The digital twin improves changeover planning, while predictive maintenance provides earlier warnings. However, the system requires continuing model monitoring, workforce training, cybersecurity updates, and periodic safety validation. The case demonstrates that no single technology solves the entire industrial problem.

Learning points

1. Safety-critical functions should remain locally enforceable.
2. Communication technology must be selected by traffic requirements, not marketing labels.
3. Machine learning requires lifecycle governance and monitoring.
4. Interoperability depends on shared meaning, not merely shared connectivity.
5. Human authority must be designed into operational procedures.

11. Professional Design Framework

A practical evaluation can follow these steps:

1. Define the mission and hazards. Identify what the system must accomplish and how failure could cause harm.
2. Classify decisions. Separate safety, control, optimization, monitoring, and administrative decisions.
3. Allocate computation. Place functions on the device, edge, site, or cloud according to latency, privacy, energy, and resilience requirements.
4. Specify data meaning. Define units, timestamps, identifiers, uncertainty, ownership, and retention.
5. Engineer degraded modes. Assume loss of sensors, communications, power, models, and external services.
6. Validate across the life cycle. Test normal, abnormal, adversarial, and maintenance conditions.
7. Assess people and society. Consider workload, accessibility, privacy, skills, employment effects, environmental impact, and accountability.
8. Plan technology replacement. Use modular interfaces and cryptographic agility to manage obsolescence.

Challenges and Limitations

Contemporary IoT systems face several persistent limitations. First, integration complexity grows as sensors, networks, models, and organizations interact. Second, data quality is often poor because sensors drift, labels are incomplete, and operating conditions change. Third, AI models may fail under distribution shift. Fourth, wireless and cloud dependence can create common points of failure. Fifth, standards and regulations evolve more slowly than commercial products. Finally, sustainability claims may ignore embodied energy, device disposal, and the social effects of automation.

Engineers should resist the assumption that newer technology is automatically better. A simple, auditable, locally controlled solution may be superior to a more sophisticated architecture that is expensive, opaque, and difficult to maintain.

Chapter Summary

Contemporary IoT in mechatronics combines physical systems with communication, computation, autonomy, and human participation. Autonomous and collaborative systems require explicit authority, safety boundaries, fault handling, and verification. Industry 5.0 broadens the objective from productivity to human centricity, sustainability, and resilience.

5G can support mobility and industrial connectivity, while TSN addresses predictable communication in suitable Ethernet environments. 6G is an emerging research direction and should not be treated as a guaranteed industrial capability. Edge computing and TinyML reduce latency and bandwidth demand; federated learning can reduce centralized data collection but introduces non-IID data, security, and governance challenges.

Swarm systems provide scalability but require collective-behavior verification. Distributed ledgers may support provenance and shared trust but cannot validate the truth of physical measurements. Energy harvesting enables low-maintenance devices when energy budgets are carefully managed. Interoperable digital twins require shared semantics, identifiers, units, provenance, and lifecycle management. Quantum-resilient security requires cryptographic inventories, agility, migration planning, and continued attention to basic cybersecurity.

The central engineering lesson is systems thinking: technologies must be selected together with safety, people, data governance, sustainability, maintainability, and organizational responsibility.

Key Terms

Autonomy: The capacity of a system to perceive, decide, and act with limited human intervention.

Collaborative robot: A robot designed to operate in a shared workspace with people under specified safety conditions.

Industry 5.0: A human-centric, sustainable, and resilient development of advanced industrial production.

Time-sensitive networking: Ethernet mechanisms that support predictable timing and bounded communication behavior.

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

TinyML: Machine learning implemented on resource-constrained embedded devices.

Federated learning: Distributed model training in which raw data remain at participating clients.

Swarm system: A population of agents whose local interactions produce coordinated collective behavior.

Distributed ledger: A replicated record maintained across multiple participants.

Energy harvesting: Conversion of ambient energy into usable electrical energy.

Digital twin: A connected computational representation of a physical asset or process.

Semantic interoperability: Consistent meaning of exchanged data across systems.

Cryptographic agility: The ability to replace cryptographic algorithms without redesigning the entire system.

Post-quantum cryptography: Cryptographic methods intended to resist attacks by classical and quantum computers.

Non-IID data: Data distributions that differ across devices, sites, or participants.

Degraded mode: A controlled operating state entered when normal resources or assumptions are unavailable.

Human centricity: Designing technology around human capability, dignity, safety, and agency.

Resilience: The capacity to withstand, recover from, and adapt to disruption.

Review Questions

Short Answer Questions

1. Distinguish an autonomous system from a connected system.
2. What are the three broad themes associated with Industry 5.0?
3. Why does the 5G label alone not guarantee deterministic industrial control?
4. State one advantage and one limitation of TinyML.
5. What is meant by cryptographic agility?

Long Answer Questions

1. Explain the engineering requirements of a safe human–robot collaborative workspace.
2. Compare cloud, edge, and federated-learning architectures for predictive maintenance.
3. Discuss how interoperability affects the design and operation of digital twins.
4. Evaluate the potential and limitations of distributed ledgers in industrial IoT.
5. Explain how Industry 5.0 changes the evaluation of automation projects.

Analytical / Application Questions

1. Design a communication architecture for a factory containing fixed motion-control equipment, mobile robots, and noncritical environmental sensors. Justify the role of TSN, private wireless, and cloud services.
2. An energy-harvesting sensor receives 0.8 mW on average but consumes 0.5 mW during sensing and 0.6 mW during a short transmission duty cycle. Explain why average power alone may be insufficient for design and propose an energy-management strategy.
3. A manufacturer wants to train a common vibration model across five factories without sharing raw data. Analyze the benefits, non-IID data risks, poisoning risks, validation process, and rollback requirements of federated learning.

Multiple Choice Questions with Answers

Question 1

Which characteristic most directly distinguishes an autonomous mechatronic system from a merely connected system?

A. It contains a wireless interface

B. It can perceive, decide, and act with limited human intervention

C. It stores data in a cloud database

D. It uses an electric actuator

Correct Answer: B. It can perceive, decide, and act with limited human intervention

Question 2

Which Industry 5.0 principle emphasizes worker agency, safety, and meaningful human participation?

A. Human centricity

B. Maximum automation

C. Data minimization

D. Cryptographic agility

Correct Answer: A. Human centricity

Question 3

What is the main purpose of time-sensitive networking in industrial Ethernet?

A. To eliminate all wireless communication

B. To provide predictable timing for selected traffic

C. To train machine-learning models locally

D. To store maintenance records immutably

Correct Answer: B. To provide predictable timing for selected traffic

Question 4

Which computation location is generally closest to the physical sensor?

A. Enterprise cloud

B. Remote data center

C. Embedded device

D. Corporate planning system

Correct Answer: C. Embedded device

Question 5

In federated learning, what is typically kept at participating clients?

A. All raw training data are necessarily sent to the server

B. Raw data, while model information is exchanged

C. Only the final production schedule

D. The physical sensor hardware

Correct Answer: B. Raw data, while model information is exchanged

Question 6

What is a principal verification challenge in swarm robotics?

A. Each robot must have unlimited computing power

B. Collective behavior may be unexpected even when individual rules appear safe

C. Swarm robots cannot communicate locally

D. Swarm systems always require a blockchain

Correct Answer: B. Collective behavior may be unexpected even when individual rules appear safe

Question 7

For photovoltaic harvesting, in the relation P_e = ηAG, what does η represent?

A. Active area

B. Incident irradiance

C. Conversion efficiency

D. Electrical resistance

Correct Answer: C. Conversion efficiency

Question 8

Which statement about a digital twin is most accurate?

A. It is always only a static 3D drawing

B. It is a connected computational representation of a physical asset or process

C. It eliminates the need for physical testing

D. It guarantees semantic interoperability automatically

Correct Answer: B. It is a connected computational representation of a physical asset or process

Question 9

Why is a ledger unable to guarantee that a sensor measurement is physically truthful?

A. Ledgers cannot store timestamps

B. A ledger can preserve an incorrect input as reliably as a correct one

C. Ledgers cannot use cryptography

D. Sensors cannot connect to networks

Correct Answer: B. A ledger can preserve an incorrect input as reliably as a correct one

Question 10

What is the purpose of cryptographic agility?

A. To increase motor speed

B. To replace cryptographic algorithms without redesigning the whole system

C. To compress all sensor data

D. To synchronize robot motion

Correct Answer: B. To replace cryptographic algorithms without redesigning the whole system