Connected Systems, IoT and Digital Transformation

Introduction

Mechatronic products increasingly operate as connected systems rather than isolated machines. A motor drive may report condition data to an edge computer; the edge computer may coordinate several machines; a cloud platform may aggregate production information; and engineering teams may use the resulting data for predictive maintenance, process optimisation and product-service innovation. This combination of embedded control, communications, data platforms and organisational change is commonly described as the Industrial Internet of Things (IIoT) and digital transformation.

Connectivity does not automatically create value. A poorly designed connected system can increase attack surfaces, produce unusable data, create dependence on vendors and interfere with safety-critical operation. The engineering task is therefore to integrate devices, networks, edge services, cloud platforms and human processes while preserving real-time performance, safety, security, maintainability and economic value.

Learning objectives

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

1. Explain the architecture and terminology of connected mechatronic systems.
2. Select suitable sensing, communication, edge and cloud functions.
3. Design telemetry, diagnostics and digital-twin workflows.
4. Evaluate interoperability, cybersecurity and OTA-update strategies.
5. Analyse lifecycle value, risks and organisational implications of digital transformation.

1. From embedded controller to connected cyber-physical system

An embedded system combines computation with a physical product. It measures variables, executes algorithms and commands actuators. A connected embedded system additionally exchanges information with other systems. A cyber-physical system (CPS) closes the loop between physical processes, computation, communication and human or organisational decisions.

A useful conceptual chain is:

physical process → sensors → embedded control → network → edge/cloud services → decisions → commands or work orders

The control loop and the information loop should not be confused. A servo controller may require deterministic millisecond-level timing and must continue operating if the internet fails. A cloud dashboard may tolerate seconds or minutes of delay and should not directly replace a safety interlock. Good architecture keeps safety and essential control local while using connectivity for supervision, optimisation and coordination.

Figure: A connected mechatronic architecture separates local control from higher-level information services.

Essential terminology

• Telemetry is the collection and transmission of measured or calculated operational data.
• OT (operational technology) controls or monitors physical processes; IT (information technology) manages information systems and business applications.
• Edge computing places processing near the data source or physical process.
• Cloud computing provides remote, scalable storage and computation through networked services.
• Interoperability is the ability of systems to exchange and use information correctly.
• IIoT platform is an integrated environment for device management, data ingestion, storage, analytics and applications.
• Digital twin is a digital representation of a physical asset or process that is linked to relevant lifecycle data. It is more than a static 3D model.
• OTA update means delivering firmware or software updates over a communication network.

2. Architecture and allocation of functions

A practical architecture is often divided into four layers.

Physical and embedded layers

Sensors measure position, temperature, pressure, vibration, current or process quality. Actuators alter motion, flow, force or energy. The embedded controller performs sampling, filtering, control, fault detection and local data preparation. Important design questions include sensor accuracy, sampling frequency, timestamp quality, memory capacity and behaviour during communication loss.

Connectivity and edge layer

Industrial Ethernet, fieldbuses, Wi-Fi, cellular networks and low-power wireless technologies serve different requirements. Selection depends on latency, determinism, range, bandwidth, mobility, electromagnetic environment and security. A gateway may translate protocols, authenticate devices, buffer data, filter events and execute analytics locally.

Edge processing is valuable when latency, privacy, reliability or bandwidth matters. For example, a vibration feature can be calculated at the machine and only the feature, alarm and selected waveform need to be transmitted. Edge autonomy also permits continued operation during cloud outages.

Cloud and enterprise layer

Cloud services commonly provide device registries, message brokers, time-series databases, dashboards, fleet management, machine-learning services and APIs. Enterprise integration connects operational data to manufacturing execution systems (MES), computerised maintenance management systems (CMMS), enterprise resource planning (ERP) and supply-chain systems.

A good allocation principle is: safety-critical and hard real-time functions remain local; latency-tolerant aggregation and fleet analytics may be placed at the edge or cloud; business decisions are integrated with human approval and established work processes.

3. Telemetry engineering

Telemetry is not simply “collect everything.” Data must have a purpose, a defined meaning and a responsible owner. A telemetry specification should identify the variable, unit, sensor location, timestamp source, sampling or event rule, quality flag, retention period and intended use.

For a sampled signal, the sampling frequency f_s must be sufficiently high for the phenomenon being measured. Under ideal sampling conditions, a band-limited signal with maximum frequency f_max requires:

This is the Nyquist condition. It assumes appropriate anti-alias filtering and a signal genuinely limited in bandwidth. In practice, engineers select a higher rate to allow filter transition bands and transient analysis. A 100 Hz vibration component therefore cannot be reliably represented by sampling at exactly 200 Hz without practical filtering margin.

Data volume can be estimated as:

where D is uncompressed data volume in bits, n is the number of channels, f_s is samples per second, b is bits per sample and T is duration in seconds. The equation assumes continuous sampling and no protocol overhead, compression or event reduction.

Worked example. Eight channels sampled at 1 kHz with 16 bits per sample for 24 hours produce:

D = 8 × 1000 × 16 × 86400 = 11,059,200,000 bits, or approximately 1.382 GB using 8 bits per byte. Network headers, metadata, retransmissions and storage indexes increase the actual requirement. A design might therefore transmit statistical features every second and retain raw waveforms only when an anomaly occurs.

Data quality requires more than numerical values. Missing, stale, duplicated, out-of-range or implausible data should be marked with quality metadata. Time synchronisation is especially important when correlating measurements from multiple devices. A timestamp should identify whether it represents sensor acquisition time, gateway receipt time or cloud ingestion time.

4. Edge analytics, diagnostics and AI

Diagnostics can be organised into three levels:

1. Detection: determining that behaviour differs from an expected condition.
2. Isolation: identifying the likely subsystem or failure mode.
3. Prognosis: estimating how degradation may evolve and supporting maintenance timing.

Thresholds are transparent and useful for known failure modes. Model-based diagnostics compare measured behaviour with a physical or control model. Data-driven methods, including anomaly detection and classification, can identify patterns but depend on representative data and careful validation.

AI is therefore an engineering aid, not an automatic substitute for verification. A model may fail under changed loads, sensor replacement, seasonal conditions or previously unseen faults. In safety-related applications, an AI recommendation should not bypass certified protection functions. Engineers should record training data provenance, performance by operating condition, false-alarm costs and a human response procedure.

A useful operational metric is availability:

where MTBF is mean time between failures and MTTR is mean time to repair. The relationship assumes a repairable system, stable average rates and a consistent definition of failure and repair. If MTBF is 900 hours and MTTR is 6 hours, A = 900 ÷ 906 ≈ 0.9934, or 99.34%. Connectivity can improve effective availability by reducing diagnosis and repair time, but it cannot compensate for poor mechanical reliability.

5. Digital twins

A digital twin should be defined by its purpose and fidelity. A monitoring twin may contain identity, configuration, operating state and history. A predictive twin may include a degradation model. A design twin may support simulation and commissioning. These forms should not be treated as interchangeable.

A practical twin includes:

• a stable asset identity and hierarchy;
• configuration, firmware and calibration records;
• current state and historical telemetry;
• events, alarms, maintenance and work orders;
• relationships to drawings, manuals and spare parts;
• data lineage, access rights and update responsibility.

The twin must remain synchronised enough for its intended decision. If a machine changes configuration but the twin does not, analytics can generate unsafe or misleading conclusions. Model fidelity should be justified by value: a detailed physical simulation is unnecessary for a dashboard that only requires operating state.

6. Interoperability and information models

Interoperability has several dimensions:

Protocols such as MQTT are lightweight publish–subscribe mechanisms well suited to telemetry; HTTP-based APIs are common for web integration; OPC UA supports industrial information modelling, security and structured interoperability; fieldbus and real-time Ethernet technologies support deterministic machine communication. No protocol eliminates the need for a well-defined information model.

Interoperability requires naming conventions, units, engineering ranges, quality codes, timestamps, versioning and error semantics. A tag named “speed” is inadequate unless its unit, reference frame, scaling and operational meaning are specified. Open standards reduce lock-in, but gateways and adapters may still be required because legacy equipment rarely conforms to one universal model.

7. Cybersecurity, safety and governance

Connecting an embedded device changes its risk profile. Threats include unauthorised access, malware, credential theft, denial of service, manipulated telemetry, unsafe commands and supply-chain compromise. Security should be treated as a lifecycle property rather than an add-on.

Core practices include unique device identities, secure boot, signed firmware, encrypted communication, least-privilege access, network segmentation, credential rotation, vulnerability management, logging, backup and incident response. A zero-trust approach avoids assuming that a device is safe merely because it is inside a factory network.

Safety and cybersecurity interact. A cyberattack may create a hazardous motion command, while an overly aggressive security control may prevent a necessary safety update. Safety functions should have independent, fail-safe mechanisms and clearly defined states for communication loss. Risk assessment should consider both accidental failures and intentional attacks.

Governance also covers privacy and accountability. Industrial telemetry may reveal worker performance, production rates or customer information. Collection should be proportionate, access should be controlled and retention should have a documented rationale. Regulations and standards vary by jurisdiction and sector; engineers should consult applicable machinery-safety, cybersecurity, data-protection and critical-infrastructure requirements rather than assume that one framework is sufficient.

8. OTA updates and fleet management

OTA updates improve security and maintainability but introduce the possibility of fleet-wide failure. A robust process includes version control, compatibility checks, signed packages, staged deployment, health monitoring and rollback. A device should retain a known-good image and change to the new image only after integrity and compatibility checks succeed.

A deployment strategy may proceed from laboratory devices to a pilot group, then to a small production percentage and finally to the whole fleet. The rollout should define stop criteria, such as increased reboot rate, abnormal current consumption or loss of telemetry. Safety-critical firmware may require formal verification, regulatory review or on-site procedures.

Device lifecycle management also requires inventory. For every asset, an organisation should know its owner, hardware revision, software version, certificate status, support period and end-of-life plan. Unsupported devices are a material operational risk even if they still function.

9. Evaluating lifecycle value

The business case must include the complete lifecycle, not only the purchase price. Costs include sensors, installation, connectivity, platform subscriptions, integration, cybersecurity, training, data storage, maintenance and eventual retirement. Benefits may include reduced downtime, lower energy use, improved quality, reduced travel, safer maintenance and new service revenue.

A simple net-present-value model is:

where C₀ is initial cost, B_t is benefit in period t, C_t is operating cost in that period, r is the discount rate per period and N is the evaluation horizon. The model assumes cash flows can be estimated and that the discount rate is appropriate. Intangible benefits and uncertainty should be analysed separately rather than hidden in optimistic estimates.

Transformation is socio-technical. A predictive-maintenance system creates value only when technicians trust its alerts, spare parts are available, work orders are integrated and managers permit planned intervention. Training, role redesign and change management are therefore engineering deliverables, not optional communication activities.

10. Illustrative case: connected packaging line

Background. A food-packaging plant experienced intermittent conveyor stoppages. The line used legacy PLCs, local motor drives and manual maintenance records.

Problem. Failures were difficult to reproduce. Maintenance personnel replaced components after alarms, but the plant had no consistent history linking vibration, motor current, product changeovers and work orders.

Analysis. Engineers identified conveyor bearings and drive couplings as high-value targets. They installed vibration and current sensors, retained existing PLC safety functions and added an industrial gateway. The gateway timestamped data, calculated RMS vibration and transmitted events rather than continuous raw waveforms.

Approach. The system used local buffering during network outages, certificate-based device identity and a dashboard linked to the CMMS. A digital twin recorded asset identity, bearing type, firmware, alarm history and maintenance actions. A staged OTA process was tested on one gateway before wider deployment. Initially, a transparent rule-based diagnostic was used; machine learning was deferred until sufficient labelled failure data existed.

Outcome. Technicians received earlier warnings and could schedule bearing inspection during planned sanitation periods. The main benefit was not the dashboard itself but the integration of telemetry, diagnosis, work orders and technician feedback. The plant also discovered that inconsistent lubrication practice was a major contributor, demonstrating that digital evidence can expose process causes rather than merely predict component failure.

Learning points. Start with a measurable operational problem; preserve local control and safety; define data semantics; integrate with existing work; validate alerts against real outcomes; and treat cybersecurity and lifecycle support as part of the design.

11. Limitations and contemporary directions

Connected systems face bandwidth constraints, sensor drift, incomplete datasets, vendor lock-in, legacy protocols and uncertain return on investment. Cloud dependence may create outages or regulatory concerns. Digital twins can become expensive documentation projects if no decision uses their information. Predictive analytics may produce false alarms that cause alert fatigue.

Contemporary development is moving toward industrial edge AI, private 5G, time-sensitive networking, interoperable asset information models, energy-aware computing and more autonomous fleet management. These technologies are promising, but adoption should follow demonstrated requirements. A simpler, well-governed edge gateway may be superior to a sophisticated cloud architecture that cannot meet latency, security or maintenance constraints.

Best-practice design sequence

1. Define the operational decision and its required evidence.
2. Identify assets, hazards, stakeholders and lifecycle responsibilities.
3. Allocate functions between device, edge, cloud and human operators.
4. Specify data semantics, timing, quality and retention.
5. Select interoperable protocols and secure identities.
6. Prototype with representative failure and network-loss conditions.
7. Validate diagnostic performance and false-alarm consequences.
8. Plan OTA updates, vulnerability response and end-of-life support.
9. Measure technical, economic, safety and organisational outcomes.
10. Scale only after the pilot has a repeatable operating model.

Chapter Summary

Connected mechatronic systems combine physical equipment, embedded control, communication networks, edge computing, cloud services and enterprise processes. Local control must remain dependable and safe even when higher-level connectivity fails. Effective telemetry requires purpose, semantics, timing, quality metadata and appropriate data reduction. Edge analytics can reduce latency and bandwidth, while cloud platforms support fleet-scale storage and analysis.

Digital twins connect asset identity, configuration, state, history and lifecycle events, but their fidelity should match the decisions they support. Interoperability is technical, syntactic, semantic and organisational. Cybersecurity, safety, privacy, OTA updates and lifecycle support must be designed from the beginning. Finally, digital transformation succeeds only when technology improves real work and produces measurable lifecycle value.

Key Terms

Telemetry: Measurement data transmitted from an asset to another system.

Industrial Internet of Things: Connected industrial devices, systems and services that exchange operational data.

Cyber-physical system: A system integrating physical processes with computation, communication and control.

Edge computing: Processing performed near the physical data source.

Cloud computing: Network-accessible computing and storage services.

Gateway: A device that connects networks, translates protocols or performs local processing.

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

Semantic interoperability: Agreement on the meaning, units and context of data.

Digital twin: A lifecycle-linked digital representation of a physical asset or process.

Telemetry quality flag: Metadata indicating whether a value is valid, stale, missing or suspect.

Time-series database: A database optimised for values indexed by time.

Predictive maintenance: Maintenance scheduled using evidence about likely degradation or failure.

OTA update: Software or firmware delivery through a network.

Secure boot: A mechanism that permits execution only of authenticated software.

Device identity: A distinct, managed identity used to authenticate an endpoint.

MES: Manufacturing execution system coordinating production operations.

CMMS: Computerised maintenance management system for maintenance planning and records.

MTBF: Mean time between failures for a repairable asset.

MTTR: Mean time to repair an asset after failure.

Digital transformation: Organisational and operational change enabled by digital technologies.

Review Questions

Short Answer Questions

1. Distinguish a control loop from an information loop in a connected mechatronic system.
2. What information should accompany a telemetry value besides its numerical magnitude?
3. State two reasons for placing analytics at the edge.
4. Why is a digital twin more than a three-dimensional CAD model?
5. Identify two protections required for secure OTA updates.

Long Answer Questions

1. Explain a layered IIoT architecture and justify how functions should be divided between device, edge, cloud and enterprise systems.
2. Discuss the four dimensions of interoperability and explain why a common communication protocol alone is insufficient.
3. Develop a cybersecurity and safety strategy for a network-connected robotic cell.
4. Explain how telemetry, diagnostics, CMMS integration and human decision-making combine in predictive maintenance.
5. Discuss the lifecycle economic, organisational and technical risks of digital transformation.

Analytical / Application Questions

1. A machine has eight channels sampled at 1 kHz with 16-bit values for 24 hours. Calculate the ideal uncompressed data volume in gigabytes and recommend one method of reducing network traffic.
2. A repairable system has an MTBF of 900 hours and an MTTR of 6 hours. Calculate its steady-state availability and explain why connectivity might improve, but cannot guarantee, this result.
3. Design an architecture for a remote pumping station that must continue safe operation during a cloud outage. Specify local functions, edge functions, cloud functions, telemetry, security controls and OTA-update safeguards.

Multiple Choice Questions with Answers

Question 1

Which function should normally remain local to a mechatronic machine?

A. Long-term fleet benchmarking

B. Safety interlocking and hard real-time control

C. Enterprise inventory forecasting

D. Cross-site dashboard styling

Correct Answer: B. Safety interlocking and hard real-time control

Question 2

What is the main purpose of edge computing?

A. To eliminate all sensors

B. To place processing near the physical process

C. To replace every safety controller

D. To guarantee unlimited cloud storage

Correct Answer: B. To place processing near the physical process

Question 3

Eight channels sampled at 1 kHz with 16 bits per sample generate what ideal data volume in 24 hours?

A. Approximately 1.38 GB

B. Approximately 13.8 MB

C. Approximately 138 GB

D. Approximately 0.138 GB

Correct Answer: A. Approximately 1.38 GB

Question 4

Which condition is required by the ideal Nyquist sampling criterion for a highest signal frequency f_max?

A. f_s < f_max

B. f_s = f_max

C. f_s > 2f_max

D. f_s > f_max ÷ 2

Correct Answer: C. f_s > 2f_max

Question 5

Which item is a semantic interoperability requirement?

A. A physical Ethernet cable

B. A shared definition and unit for a variable

C. A larger monitor

D. A faster processor clock

Correct Answer: B. A shared definition and unit for a variable

Question 6

What distinguishes a digital twin from a static model?

A. It must always be a 3D rendering

B. It is linked to relevant lifecycle or operational data

C. It cannot contain maintenance records

D. It operates independently of the physical asset

Correct Answer: B. It is linked to relevant lifecycle or operational data

Question 7

Which OTA practice limits the impact of a defective release?

A. Deploying immediately to every device

B. Removing rollback capability

C. Using staged deployment with health monitoring

D. Sharing one password across the fleet

Correct Answer: C. Using staged deployment with health monitoring

Question 8

For a repairable system, availability is calculated as:

A. MTTR ÷ MTBF

B. MTBF ÷ (MTBF + MTTR)

C. MTBF × MTTR

D. MTBF + MTTR

Correct Answer: B. MTBF ÷ (MTBF + MTTR)

Question 9

Which statement about AI-based industrial diagnostics is most accurate?

A. It eliminates the need for validation

B. It is reliable regardless of operating conditions

C. Its performance depends on representative data and verification

D. It should directly replace certified safety functions

Correct Answer: C. Its performance depends on representative data and verification

Question 10

Which is the best initial step in an IIoT project?

A. Collect every available signal

B. Define the operational decision and required evidence

C. Purchase the largest possible cloud subscription

D. Deploy machine learning before identifying failure modes

Correct Answer: B. Define the operational decision and required evidence