10. System Integration, Networks and IoT

Introduction and Context

Mechatronic systems increasingly consist of sensors, actuators, controllers, drives, robots, supervisory computers, databases, and cloud services distributed across a machine or factory. Their value depends not only on the performance of individual components but also on their ability to exchange information reliably, promptly, and securely. System integration is therefore a central engineering activity: it transforms independent devices into a coordinated cyber-physical system.

A modern production cell may contain a programmable logic controller (PLC), variable-frequency drives, safety controllers, machine-vision cameras, condition-monitoring sensors, human–machine interfaces, and enterprise software. These devices often use different data formats, communication rates, timing assumptions, and diagnostic methods. Networks provide the communication infrastructure, while interoperability mechanisms allow devices with different origins to work together. The Industrial Internet of Things (IIoT) extends this arrangement by connecting operational technology to analytics, enterprise systems, and cloud platforms.

This chapter examines the principal technologies and design decisions involved in such integration. Particular attention is given to CAN, Modbus, industrial Ethernet, OPC UA, wireless networks, and IoT architectures. The discussion also addresses latency, determinism, edge–cloud allocation, cybersecurity, interoperability, performance evaluation, and practical deployment strategy.

Learning Objectives

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

1. Explain the role of communication networks in integrated mechatronic systems.

2. Compare CAN, Modbus, industrial Ethernet, OPC UA, and industrial wireless technologies.

3. Distinguish bandwidth, latency, jitter, determinism, availability, and reliability.

4. Design an appropriate edge–cloud architecture for an industrial application.

5. Analyse interoperability and select suitable integration mechanisms.

6. Evaluate network performance and identify common integration risks.

7. Apply cybersecurity, governance, and lifecycle principles to connected machines.

8. Recommend a communication architecture for a realistic industrial case.

From Independent Devices to a Connected System

Integration occurs at several levels. At the device level, sensors and actuators exchange signals with controllers. At the control level, PLCs, drives, robots, and safety systems coordinate machine operations. At the supervisory level, operators monitor production and alarms. At the enterprise level, manufacturing execution systems (MES), maintenance applications, and enterprise resource planning (ERP) systems use production information. IoT architectures add a further layer in which historical and real-time data support remote monitoring, optimisation, and predictive maintenance.

A useful distinction is between operational technology (OT) and information technology (IT). OT systems directly control physical processes and traditionally prioritise deterministic behaviour, availability, and safety. IT systems prioritise scalability, data sharing, flexibility, and security across large information infrastructures. Integration between the two must preserve the time-critical and safety-related properties of OT while obtaining the analytical benefits of IT.

Interoperability has at least three dimensions. Syntactic interoperability means that systems can exchange data using compatible formats. Semantic interoperability means that the meaning of the data is understood consistently; for example, all systems interpret a value as motor temperature in degrees Celsius rather than as an unspecified integer. Operational interoperability means that the connected systems can cooperate correctly under real process conditions, including timing, alarms, access control, and failure recovery.

Communication Requirements in Mechatronics

Network selection should begin with the process rather than with a favourite protocol. Important requirements include message size, update rate, maximum permissible delay, allowable jitter, number of nodes, physical distance, electromagnetic environment, fault tolerance, safety requirements, installation cost, and future expansion.

Latency is the time between transmitting information and its successful use by the receiving application. Jitter is the variation in latency. A system can have low average latency but unacceptable jitter if some messages occasionally arrive too late. Determinism describes the predictability of communication timing. Bandwidth is the maximum data-carrying capacity, whereas throughput is the useful data successfully transferred. Availability expresses the proportion of time that a service is operational, and reliability concerns the probability of performing correctly over a specified period.

Control loops impose particularly strict requirements. If a servo controller samples position every millisecond, communication delay and jitter must be small relative to the loop period. A cloud connection may provide excellent bandwidth but cannot normally guarantee the deterministic timing required for motion control. Consequently, time-critical control is generally retained locally, while supervisory and analytical functions may be distributed to edge or cloud systems.

Controller Area Network (CAN)

CAN was developed for robust real-time communication among electronic control units. It uses a broadcast bus in which messages are identified by an identifier rather than by a destination address. All nodes observe the bus, but each node accepts only relevant messages. Arbitration is non-destructive: when several nodes transmit simultaneously, the message with the highest priority identifier continues without corrupting the frame.

CAN is attractive in mechatronics because it provides error detection, acknowledgement, automatic retransmission, and fault confinement. A faulty node can be prevented from disturbing the entire network. Its short messages and priority-based arbitration make it suitable for distributed control, mobile machinery, automotive subsystems, motor controllers, and embedded instrumentation.

CANopen adds an application-layer framework containing object dictionaries, device profiles, process data objects, service data objects, and network-management functions. It reduces the integration effort when devices conform to common profiles. CAN FD extends the data field and can provide higher payload efficiency, although all devices and infrastructure must support the selected mode.

CAN is not a universal solution. Bus loading, cable length, termination, electromagnetic interference, and priority assignment influence performance. A high-priority message can delay lower-priority traffic. Designers must calculate worst-case response time rather than rely only on average traffic. CAN is also less suitable for high-volume data such as video or large historical records.

Modbus and Its Industrial Role

Modbus is a simple and widely implemented application protocol for reading and writing coils, discrete inputs, input registers, and holding registers. Modbus RTU commonly operates over serial RS-485, while Modbus TCP uses Ethernet and TCP/IP. The protocol follows a client–server model in which a client requests data or an operation from a server.

The principal strength of Modbus is simplicity and broad device availability. It is often used for meters, drives, remote I/O, building systems, and legacy equipment. RS-485 supports multiple devices on a shared physical link, but installation requires attention to termination, polarity, grounding, shielding, and addressing. Modbus TCP simplifies physical networking but does not by itself provide deterministic real-time behaviour, rich semantic models, or strong security.

A frequent integration problem is inconsistent register interpretation. One manufacturer may store a 32-bit floating-point value in two registers with one word order, while another uses a different order. Scaling, signedness, units, and update rates must therefore be documented explicitly. Modbus is effective as a practical interface, but additional gateways, data models, and security controls may be required for a modern connected architecture.

Industrial Ethernet

Industrial Ethernet uses Ethernet-based communication in environments requiring higher bandwidth, longer distances, diagnostics, and integration with standard IT infrastructure. Examples include PROFINET, EtherNet/IP, EtherCAT, POWERLINK, and Sercos. Although they share Ethernet technologies, their application layers and real-time mechanisms differ.

Ordinary office Ethernet uses contention management and switching, but industrial control may require scheduled or prioritised traffic, synchronised clocks, redundancy, and specialised hardware. EtherCAT, for example, processes frames efficiently as they pass through slave devices. Time-sensitive networking (TSN) extends Ethernet with mechanisms for time synchronisation, traffic scheduling, and bounded transmission behaviour. Such technologies can support the convergence of control, safety, and information traffic, provided that the complete system is engineered accordingly.

Industrial Ethernet offers high bandwidth and flexible topology, but it does not automatically guarantee performance. Switch configuration, traffic segregation, cable quality, electromagnetic compatibility, multicast behaviour, and network management are significant. VLANs, quality-of-service rules, redundancy protocols, and network monitoring should be applied deliberately rather than added after commissioning.

OPC UA as an Interoperability Layer

OPC Unified Architecture (OPC UA) is a platform-independent framework for secure and semantically meaningful industrial information exchange. Unlike a simple register protocol, OPC UA can represent objects, variables, methods, events, alarms, relationships, units, and metadata. A client can discover information from a server rather than relying solely on undocumented memory addresses.

OPC UA supports client–server communication and publish–subscribe models. It includes authentication, encryption, signing, certificates, role-based access, and auditing mechanisms. Its information modelling capability is particularly valuable when integrating equipment from different vendors or connecting plant systems to MES and analytics platforms.

OPC UA does not replace every field-level network. A motion system may still use EtherCAT or another real-time network for its local loop, while an OPC UA server exposes selected machine states, production counts, alarms, and diagnostics to higher-level systems. The important architectural principle is separation of functions: field networks handle deterministic control, and OPC UA provides interoperable information exchange above them.

Wireless Industrial Networks

Wireless communication is useful where cabling is expensive, moving equipment is involved, or temporary and mobile sensors are required. Technologies include industrial Wi-Fi, Bluetooth Low Energy, WirelessHART, ISA100.11a, private cellular networks, and low-power wide-area networks.

Wireless selection depends on range, power consumption, data rate, mobility, interference, latency, and security. Industrial Wi-Fi can support high data rates and mobile terminals but requires careful radio planning. WirelessHART and ISA100.11a are designed for industrial sensing and commonly use mesh techniques, channel management, and redundancy. Private 5G can offer managed mobility, broad coverage, and quality-of-service mechanisms, although its cost and engineering complexity must be justified.

Wireless networks are vulnerable to obstruction, interference, fading, congestion, and changing radio conditions. A wireless link should not be treated as equivalent to a physical wire merely because its nominal data rate is high. Applications must define acceptable packet loss, retry behaviour, buffering, and safe fallback states. Safety-critical functions should use appropriately certified architectures and should not depend casually on an unverified wireless path.

IoT and Edge–Cloud Architectures

An industrial IoT architecture normally contains sensing and actuation, local control, edge processing, communication and messaging, data platforms, applications, and governance. Data may move upward from a sensor to a gateway, edge computer, plant server, and cloud service. Commands and configuration move in the opposite direction, subject to authorisation and safety constraints.

Edge computing places processing close to the equipment. It is appropriate when decisions require low latency, data must remain on site, connectivity is intermittent, or raw data volume is large. Cloud computing is useful for long-term storage, fleet-level comparison, large-scale analytics, model training, remote collaboration, and elastic computing capacity. The correct allocation is rarely either entirely edge or entirely cloud.

A practical allocation is as follows:

• The controller performs closed-loop control and immediate interlocking.
• An edge node filters signals, detects events, aggregates data, and executes low-latency diagnostics.
• A plant server provides dashboards, historian functions, and local coordination.
• The cloud stores selected historical data, compares multiple assets, trains predictive models, and supports enterprise applications.

Publish–subscribe messaging, often implemented using MQTT or similar middleware, reduces direct point-to-point dependencies. However, message brokers, topic structures, quality-of-service settings, retained messages, timestamps, and identity management must be designed carefully. Data governance is essential: collecting everything without a clear purpose creates storage, security, and interpretation problems.

A Framework for Integration Design

A disciplined integration process begins with functional and performance requirements. The engineer should identify what must be controlled, monitored, recorded, diagnosed, or optimised. Each data item should have an owner, source, unit, update rate, timestamp, quality status, and permitted users.

The next step is architectural decomposition. Devices are grouped into suitable domains: safety, motion, basic control, supervision, enterprise information, and external analytics. Communication boundaries are then selected. A gateway may translate CAN or Modbus into OPC UA, but translation should preserve timestamps, quality indicators, units, alarms, and failure states rather than merely copying numerical values.

Network loading should be estimated before installation. For periodic messages, approximate offered traffic can be calculated from message size and transmission frequency, with allowances for protocol overhead and retransmissions. Designers should examine worst-case bursts, not only average rates. Response-time analysis should include device processing, queueing, transmission, gateway conversion, and application execution.

Verification should include normal operation, overload, cable disconnection, device restart, time synchronisation failure, invalid data, cyberattack, and loss of external connectivity. A connected system is not complete when data appear on a dashboard; it is complete when failures produce safe, understandable, and recoverable behaviour.

Applied Case: Connected Condition Monitoring in a Packaging Line

Background

The following illustrative case concerns a packaging line containing conveyors, servo drives, pneumatic actuators, a PLC, a safety controller, barcode readers, and a supervisory workstation. The plant wishes to reduce unplanned stoppages caused by conveyor bearings and drive overheating.

Situation and Problem

The drives communicate with the PLC through an industrial Ethernet network. Several older temperature sensors expose values through Modbus RTU, while a vibration sensor manufacturer proposes a wireless interface. Production data are recorded locally, but maintenance personnel have no historical correlation between vibration, temperature, product rate, and failures. Management requests cloud dashboards and predictive-maintenance alerts.

A direct connection of every device to the cloud would create several problems: control traffic would depend on external connectivity, raw vibration data would consume substantial bandwidth, and the security boundary of the control network would be weakened. The engineering team therefore defines three classes of information. Servo commands and interlocks are time-critical; temperatures, speeds, and alarms are supervisory; vibration features and maintenance history are analytical.

Analysis and Architecture

The existing industrial Ethernet network remains responsible for PLC and drive coordination. Modbus RTU sensors are connected through a gateway that converts register values into named variables with units, scaling, timestamps, and quality flags. A local edge computer receives vibration samples through the wireless network, calculates statistical features such as root-mean-square vibration and crest factor, and retains short-term raw data for diagnosis.

The edge computer publishes selected information through an MQTT broker located in a demilitarised industrial zone. An OPC UA server exposes machine states, alarms, and maintenance-relevant variables to the plant supervisory system. Only approved aggregates and event records are sent to the cloud. The cloud platform compares similar machines and provides long-term trend analysis. No cloud command is permitted to bypass the PLC or safety controller.

Performance and Security Measures

The team measures sensor-to-edge delay, edge processing time, broker delay, dashboard refresh time, packet loss, and CPU utilisation. Alarm rules are executed at the edge so that loss of cloud connectivity does not prevent local notification. Network segmentation separates the control network, plant services, and external connection. Certificates and role-based access protect OPC UA and MQTT services; unused ports are disabled; firmware and gateway configurations are backed up.

Outcome

The resulting architecture supports local control, meaningful data integration, and remote analysis without making production dependent on the cloud. Maintenance staff receive local alerts and historical trends, while the organisation obtains fleet-level information. The principal benefit does not arise from any single protocol. It results from assigning each communication technology to the function for which it is technically and economically appropriate.

Learning Points

1. Time-critical control should remain within a deterministic local domain.

2. Protocol conversion must preserve semantics, quality, and time information.

3. Edge processing can reduce data volume and preserve operation during disconnection.

4. Cloud analytics is most valuable when based on well-defined, trustworthy data.

5. Integration architecture and cybersecurity must be designed together.

Challenges, Risks, and Common Mistakes

Legacy equipment often lacks documentation, meaningful timestamps, or secure authentication. Protocol gateways may conceal errors by converting invalid or stale values into apparently normal numbers. Another common mistake is to assume that a high nominal bandwidth guarantees low control delay. Queueing, retransmission, gateway processing, and overloaded servers can dominate performance.

Point-to-point integrations create brittle systems in which replacing one device requires changes across many applications. Poor naming conventions and inconsistent units undermine semantic interoperability. Excessive polling increases network load, whereas insufficient polling causes stale information. Wireless systems may be installed without a radio survey, and cloud projects may collect large quantities of data without specifying the decisions the data should support.

Cybersecurity risks include unauthorised access, malware propagation, credential theft, denial of service, manipulation of process values, and insecure remote maintenance. Safety and cybersecurity are related but not identical. A secure system can still be unsafe if its control logic is wrong, and a safety function cannot be justified merely because a network is encrypted.

Governance, Ethics, and Professional Responsibility

Connected industrial systems generate information about equipment, workers, production rates, and maintenance behaviour. Data collection should have a legitimate purpose, defined retention period, access policy, and responsible owner. Worker monitoring should be transparent and proportionate. Engineers must avoid presenting a predictive-maintenance model as certain when it produces probabilistic results.

Professional practice requires traceable configuration, validated changes, controlled access, documented assumptions, and tested recovery procedures. Relevant organisational and sectoral requirements may include functional-safety standards, industrial cybersecurity guidance, data-protection obligations, electromagnetic-compatibility rules, and machinery regulations. Compliance does not replace engineering judgement, but it establishes minimum expectations for risk control and accountability.

Current Developments and Best Practices

Current industrial architectures increasingly combine OPC UA information models, MQTT-style event distribution, edge analytics, containerised applications, digital twins, private cellular networks, and time-sensitive networking. Artificial intelligence is being applied to anomaly detection, remaining-useful-life estimation, vision inspection, and adaptive scheduling. Nevertheless, AI performance depends on representative data, reliable labels, model monitoring, and safe handling of uncertainty.

Best practice is to:

• define requirements before selecting protocols;
• separate safety, control, supervisory, and analytical functions;
• use standard information models and consistent naming;
• measure worst-case latency, jitter, loading, and recovery time;
• design segmentation, authentication, patching, and backups from the beginning;
• retain local safe operation during cloud or network failure;
• test gateways and data quality under abnormal conditions;
• document ownership, units, timestamps, quality codes, and lifecycle responsibilities;
• introduce new connectivity incrementally through a monitored pilot.

Chapter Summary

System integration connects distributed mechatronic devices into a coordinated cyber-physical system. CAN provides robust priority-based embedded communication; Modbus offers simple and widely available register-based exchange; industrial Ethernet supports higher bandwidth and, in specialised forms, deterministic control; OPC UA supplies secure, semantic, platform-independent interoperability; and wireless technologies enable mobile and difficult-to-wire applications. IoT architectures distribute functions among devices, edge nodes, plant systems, and clouds.

Successful design depends on matching communication technology to process requirements. Latency, jitter, determinism, bandwidth, availability, data quality, and recovery behaviour must be analysed together. Time-critical control should remain local, while edge and cloud resources can support filtering, diagnostics, storage, optimisation, and fleet analytics. Interoperability requires semantic models as well as physical connectivity. Cybersecurity, governance, safety, and lifecycle management are integral engineering concerns rather than additions made after installation.

Key Terms

CAN: A robust multi-master embedded network using identifier-based message arbitration.

CANopen: An application-layer framework that standardises device objects and services over CAN.

Modbus: A simple industrial protocol for reading and writing discrete and register-based data.

Industrial Ethernet: Ethernet-based networking adapted for industrial communication and control.

OPC UA: A secure, platform-independent framework for semantic industrial information exchange.

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

Latency: The time taken for information to travel from source to application or destination.

Jitter: Variation in communication delay or timing.

Determinism: Predictability of system response and communication timing.

Throughput: The useful amount of data successfully transferred per unit time.

Edge computing: Processing performed near the physical equipment or data source.

Cloud computing: On-demand computing and storage delivered through a remote network infrastructure.

MQTT: A lightweight publish–subscribe messaging protocol commonly used in IoT systems.

Gateway: A device or software component that connects networks or translates protocols.

Semantic interoperability: Consistent interpretation of the meaning, units, and context of data.

Network segmentation: Dividing a network into controlled zones to limit traffic and risk.

Quality of service: Rules that assign communication treatment or priority to different traffic classes.

Time-sensitive networking: Ethernet mechanisms supporting synchronised and bounded communication timing.

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

Industrial IoT: Connected sensing, control, computing, and analytics applied to industrial operations.

Review Questions

Short Answer Questions

1. Distinguish latency from jitter in an industrial network.

2. Why is CAN arbitration described as non-destructive?

3. State two limitations of Modbus as a modern integration technology.

4. What is the principal contribution of OPC UA to interoperability?

5. Give two reasons for processing data at the edge.

Long Answer Questions

1. Compare CAN, Modbus, industrial Ethernet, and OPC UA with respect to communication model, typical use, timing, and integration capability.

2. Explain how an industrial IoT architecture can divide responsibilities among a controller, edge computer, plant server, and cloud platform.

3. Discuss the design factors that determine whether a wireless network is appropriate for an industrial application.

4. Describe a systematic process for designing, verifying, and commissioning a connected mechatronic system.

5. Explain why cybersecurity, safety, data governance, and interoperability must be treated as integrated engineering concerns.

Analytical / Application Questions

1. A robotic assembly cell requires sub-millisecond motion coordination, machine alarms, production reporting, and monthly fleet analytics. Propose a layered communication architecture and justify the placement of each function.

2. A Modbus temperature sensor occasionally reports plausible but stale values after a gateway restart. Analyse the likely integration weakness and specify data-quality and recovery mechanisms.

3. A factory plans to send all raw vibration data directly to the cloud. Evaluate this proposal and recommend an alternative edge–cloud allocation, including performance, cost, reliability, and cybersecurity considerations.

22. Multiple Choice Questions with Answers

Question 1

Which property of CAN arbitration allows the highest-priority message to continue without corrupting the frame?

A. Token passing

B. Non-destructive arbitration

C. Store-and-forward switching

D. Frequency hopping

Correct Answer: B. Non-destructive arbitration

Question 2

A major integration risk when using Modbus registers from different vendors is variation in:

A. Laser wavelength

B. Register scaling and word order

C. Cloud storage capacity

D. Mechanical bearing clearance

Correct Answer: B. Register scaling and word order

Question 3

Which technology is primarily intended to provide semantic, secure, and platform-independent industrial information exchange?

A. RS-485

B. CAN FD

C. OPC UA

D. PWM

Correct Answer: C. OPC UA

Question 4

Why should a motion-control loop generally not depend directly on a cloud connection?

A. Cloud systems cannot store historical data

B. Cloud connections cannot provide the required deterministic local timing

C. Cloud systems use only analogue signals

D. Motion controllers cannot communicate digitally

Correct Answer: B. Cloud connections cannot provide the required deterministic local timing

Question 5

In the illustrative packaging-line case, why were vibration features calculated at the edge?

A. To eliminate the need for sensors

B. To increase raw data transmission to the cloud

C. To reduce data volume and preserve local diagnostic capability

D. To bypass the PLC safety controller

Correct Answer: C. To reduce data volume and preserve local diagnostic capability

Question 6

Which measure most directly describes variation in message delivery time?

A. Bandwidth

B. Jitter

C. Availability

D. Payload

Correct Answer: B. Jitter

Question 7

What is the main purpose of network segmentation in an industrial architecture?

A. To make all devices use identical hardware

B. To limit traffic movement and reduce security exposure

C. To remove the need for authentication

D. To guarantee unlimited bandwidth

Correct Answer: B. To limit traffic movement and reduce security exposure

Question 8

Which statement best describes semantic interoperability?

A. Two devices use the same cable connector

B. Two networks have equal bandwidth

C. Two systems interpret exchanged data consistently, including units and meaning

D. Two controllers operate at different sampling rates

Correct Answer: C. Two systems interpret exchanged data consistently, including units and meaning

Question 9

A wireless sensor is installed in a factory without a radio survey and experiences intermittent packet loss. Which factor was most clearly neglected?

A. Mechanical stiffness

B. Radio interference and propagation conditions

C. OPC UA information modelling

D. PLC programming language

Correct Answer: B. Radio interference and propagation conditions

Question 10

Which allocation is most appropriate for a cloud platform in a connected factory?

A. Executing every safety interlock

B. Closing a sub-millisecond servo loop

C. Performing long-term fleet analytics and model training

D. Replacing all local network diagnostics

Correct Answer: C. Performing long-term fleet analytics and model training