Energy, Thermal and Sustainability Applications

Introduction and context

Mechanical systems are major users and transformers of energy. Motors drive pumps and compressors; boilers convert fuel into steam; turbines convert fluid energy into shaft work; heating, ventilation and air-conditioning (HVAC) systems maintain indoor conditions; and process equipment transfers heat and mass. These systems are increasingly instrumented with connected sensors, supervisory control and data acquisition (SCADA), building-management systems, industrial communication networks and cloud or edge analytics.

Artificial intelligence (AI) and the Internet of Things (IoT) are valuable not because they replace thermodynamics or engineering judgment, but because they make energy behaviour more observable, predictable and controllable. An IoT system can collect temperatures, pressures, flow rates, electrical power, valve positions, vibration and emissions data. AI can then identify abnormal consumption, forecast demand, estimate unmeasured variables, optimise setpoints and support maintenance decisions.

The engineering objective is broader than reducing an energy bill. A sound intervention should consider safety, product quality, reliability, greenhouse-gas emissions, water use, material consumption, capital cost, occupant comfort and equipment life. Energy efficiency is therefore a technical and organisational problem involving measurement, modelling, control and decision-making.

Learning objectives

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

1. Explain how AI–IoT architectures support energy and thermal management.
2. Apply energy, power, efficiency and heat-transfer relationships to monitored systems.
3. Select useful performance indicators for HVAC, boilers, turbines and fluid systems.
4. Distinguish anomaly detection, forecasting, diagnosis and optimisation.
5. Evaluate energy-efficiency measures using lifecycle and sustainability criteria.
6. Identify data, safety, cybersecurity, ethical and governance risks.
7. Develop a practical monitoring and improvement plan for a mechanical system.

1. Energy and sustainability foundations

1.1 Energy, power and efficiency

Energy is the capacity to produce work or transfer heat, measured in joules (J). Power is the rate of energy transfer. For a time interval, average power is

where P is average power in watts (W), E is energy in joules, and ∆t is elapsed time in seconds. Electrical meters commonly report energy in kilowatt-hours (kWh): 1 kWh = 3.6 MJ.

For a useful output and an energy input, first-law efficiency is

where η is dimensionless and normally expressed as a percentage. The equation applies to a defined boundary and time period. The input and output must be measured on compatible energy bases. For example, boiler efficiency may compare useful heat delivered to water with the chemical energy supplied by fuel, whereas motor efficiency compares mechanical shaft output with electrical input.

The first law accounts for conservation of energy but not the quality of energy. The second law introduces irreversibility and exergy: electrical work is highly convertible into useful work, while low-temperature waste heat has less ability to produce work. A system can therefore have acceptable first-law efficiency while still destroying substantial opportunities for useful work through large temperature differences, throttling, friction or combustion irreversibility.

1.2 Energy balance for a control volume

For a steady-flow device, neglecting changes in kinetic and potential energy when justified, the general energy balance is

Here Q̇ is net heat transfer into the control volume (W), Ẇ_s is shaft work leaving the control volume (W), ṁ is mass flow rate (kg/s), and h is specific enthalpy (J/kg). The equation assumes steady operation; transient storage must be added when equipment is warming, cooling or filling. Sensor data should be interpreted within the same system boundary used for the balance.

For sensible heating or cooling of a fluid with approximately constant specific heat,

where c_p is specific heat capacity (J/(kg·K)) and temperatures are in °C or K when only a difference is used. This relationship is applicable when phase change is absent and c_p varies little over the temperature range. For water, a practical approximation is c_p ≈ 4.18 kJ/(kg·K) near ordinary temperatures.

1.3 Sustainability and resource productivity

Sustainability analysis considers environmental, economic and social performance over a life cycle. Important indicators include:

• energy intensity: energy consumed per unit of product, mass, area or service;
• carbon intensity: greenhouse-gas emissions per unit of product or energy;
• water intensity: water consumed per unit of product;
• material productivity: useful output per unit of material input;
• availability and reliability: the proportion of required time that equipment can perform its function;
• lifecycle cost: acquisition, installation, energy, maintenance, downtime, disposal and residual-value costs.

A reduction in electricity use does not automatically guarantee lower total environmental impact. For example, a control strategy that reduces compressor energy but shortens equipment life, increases refrigerant leakage or compromises indoor air quality may be undesirable. Sustainability decisions should therefore use multiple indicators and explicitly state system boundaries.

2. AI–IoT architecture for energy systems

A useful architecture has five interacting layers.

1. Physical layer: meters, temperature and pressure sensors, flow meters, power analysers, gas analysers, vibration sensors, actuators and equipment.
2. Connectivity layer: wired or wireless networks, industrial Ethernet, field buses and secure gateways.
3. Data layer: time synchronisation, historians, databases, asset metadata, calibration records and data-quality flags.
4. Analytics layer: dashboards, rules, statistical models, machine learning, digital twins and optimisation routines.
5. Action and governance layer: alarms, operator workflows, control commands, maintenance tickets, access control and audit records.

The quality of the analytics is limited by the quality of the measurement system. A temperature sensor with a systematic bias can make a valid model appear inaccurate. Missing timestamps, sensor drift, duplicated records and changes in operating mode can produce false alarms. Each measurement should therefore have a tag, unit, location, sampling interval, calibration status and quality code.

2.1 From data to engineering action

A practical analytics workflow is:

1. define the engineering question and system boundary;
2. identify energy inputs, useful outputs and important disturbances;
3. select sensors and verify measurement uncertainty;
4. establish a baseline under comparable operating conditions;
5. clean, synchronise and contextualise the data;
6. develop a transparent diagnostic or predictive model;
7. validate the model against unseen data and engineering constraints;
8. recommend or implement an intervention;
9. verify savings using an agreed baseline;
10. continuously monitor performance and revise the model.

AI tasks should be matched to decisions. Forecasting predicts future load, temperature or energy use. Anomaly detection identifies behaviour that differs from normal operation. Fault detection and diagnosis links deviations to probable causes. Optimisation selects operating conditions subject to constraints. Predictive maintenance estimates failure risk or remaining useful life. These functions are related but not interchangeable: an anomaly detector may identify a problem without identifying its cause.

2.2 Baselines and normalisation

Energy use depends on production, weather, occupancy, schedule and operating mode. Comparing raw monthly energy values can therefore be misleading. A simple baseline may be represented as

where E_baseline is expected energy over a selected interval, X_i are explanatory variables such as production or heating degree-hours, and β_i are fitted coefficients. The model is useful only when the variables are measured consistently, the relationship is appropriate for the operating range and important nonlinearities or regime changes are addressed.

Savings for a reporting period can be estimated by

The adjusted baseline represents what the system would have consumed under current conditions without the intervention. Savings verification must account for uncertainty, changed production and changes in weather or occupancy. A model should not be used to claim savings outside the range in which it was validated.

3. HVAC applications

HVAC systems combine thermodynamics, fluid flow, electrical drives, control and human comfort. IoT monitoring commonly includes supply and return-air temperature, humidity, carbon-dioxide concentration, outdoor conditions, fan speed, valve position, filter pressure drop, chilled-water temperatures, flow and electrical power.

For an air-handling unit, sensible cooling or heating can be approximated by

where ṁ_air is air mass flow rate and the temperature difference is defined according to the cooling or heating direction. Total cooling load may also include latent heat associated with moisture removal; a sensible-only calculation is insufficient when humidity control is important.

A chiller coefficient of performance is

where cooling capacity and input power must use the same power units. A higher COP means more cooling per unit of input power, but the value depends on chilled-water temperature, condenser conditions, part-load operation and auxiliary pumps and fans. A plant-level metric should include those auxiliaries rather than evaluating only the compressor.

AI–IoT opportunities include:

• detecting simultaneous heating and cooling caused by faulty valves or control logic;
• predicting occupancy and pre-conditioning spaces only when required;
• identifying fouled coils from changes in temperature approach and pressure drop;
• optimising chilled-water temperature, condenser-water temperature and variable-speed drives;
• detecting abnormal fan or pump power at a given flow;
• forecasting indoor carbon dioxide and maintaining ventilation within health and comfort requirements.

Optimisation must respect constraints. Energy reduction is not acceptable if it causes unsafe temperatures, excessive humidity, inadequate ventilation, condensation, microbial growth or unacceptable occupant conditions. Human override and fail-safe local control remain essential.

4. Boilers, steam and thermal processes

A boiler converts fuel energy into useful thermal energy in water or steam. Key measurements include fuel flow, combustion-air flow, oxygen and carbon-monoxide concentration, feedwater temperature, steam pressure and flow, flue-gas temperature, blowdown rate and electrical power for auxiliaries.

A direct boiler efficiency can be estimated as

where ṁ_steam is steam mass flow, h_steam and h_feedwater are specific enthalpies, ṁ_fuel is fuel mass flow and LHV is the lower heating value of the fuel. This direct method requires reliable flow, enthalpy and fuel measurements. It is distinct from an indirect method that estimates losses through flue gas, radiation, blowdown and incomplete combustion.

IoT analytics can detect burner instability, excess air, steam leaks, failed traps, poor insulation, abnormal blowdown and declining heat-transfer performance. Oxygen-trim control can improve combustion, but excessive air carries hot gases out of the stack and too little air can increase carbon monoxide and soot. AI recommendations must therefore be bounded by combustion-safety rules and validated by qualified personnel.

Condensing boilers illustrate a sustainability trade-off. Lower return-water temperature can recover latent heat from water vapour in flue gas, improving efficiency, but condensation requires suitable materials, drainage and control. A model trained under non-condensing operation may not predict the new regime correctly.

5. Turbines, pumps and fluid systems

For a turbine, the ideal specific work under an adiabatic steady-flow approximation is related to enthalpy drop. Actual turbine efficiency may be defined as

where the denominator uses the isentropic outlet state at the same exit pressure. The definition applies to a turbine stage or unit only when inlet and outlet states and the reference process are clearly specified.

For a pump, hydraulic power is

where ρ is fluid density (kg/m³), g is gravitational acceleration (m/s²), Q is volumetric flow (m³/s), and H is pump head (m). Pump input power is higher than hydraulic power because of motor, drive and pump losses. Overall efficiency is

For geometrically similar centrifugal pumps at constant impeller diameter, approximate affinity relationships are

where N is rotational speed. These relationships are approximate and require similar fluid properties, operating regions and impeller geometry. They do not justify unrestricted speed reduction: minimum flow, system resistance, motor cooling, cavitation margin and process requirements must be checked.

AI–IoT monitoring can combine pressure, flow, speed, current and vibration to detect cavitation, clogging, bearing degradation, valve misoperation and operation far from the best-efficiency point. A useful performance map is more informative than a single alarm threshold because pump efficiency changes with flow and head.

6. Emissions reduction and carbon-aware operation

Emissions analytics begins with a transparent inventory. Direct emissions may result from fuel combustion or refrigerant leakage; indirect emissions may arise from purchased electricity, steam or chilled water. For fuel use, a simplified estimate is

where m_CO₂ is carbon-dioxide mass, E_fuel is fuel energy or quantity on the same basis as the factor, and EF_CO₂ is the applicable emissions factor. The factor must specify fuel type, heating-value basis, geography and whether it represents carbon dioxide only or carbon-dioxide equivalent.

AI can reduce emissions by forecasting demand, shifting flexible loads toward lower-carbon electricity periods, reducing combustion excess air, improving heat recovery and identifying leaks. However, carbon-aware scheduling should not compromise equipment protection, production quality or grid stability. Emissions factors can change over time, and embodied emissions from new sensors, batteries, servers and replacement equipment should be considered in lifecycle assessments.

7. Lifecycle assessment and economic decisions

Lifecycle assessment examines impacts from raw-material extraction through manufacture, transport, operation, maintenance and end-of-life. A simplified lifecycle energy model is

where each term must be defined consistently and time-indexed quantities are summed over the assessment period. A lifecycle carbon model follows the same structure but applies appropriate emissions factors to each activity.

Economic evaluation may use simple payback, net present value or lifecycle cost. Simple payback is

It ignores discounting, escalation, degradation, risk and residual value. For a more complete analysis, net present value is

where C₀ is initial cost, B_t is benefit in year t, C_t is operating or maintenance cost, r is discount rate, n is study life and S_n is residual value. Cash flows must be stated in either real or nominal terms consistently.

A sensor project may have a short financial payback but poor sustainability if frequent battery replacement, electronic waste or cybersecurity infrastructure is ignored. Conversely, a measure with a longer payback may be justified by reliability, safety, compliance or emissions objectives.

8. Illustrative applied case: university central utility plant

Background

This illustrative case concerns a university campus with two electric chillers, cooling towers, variable-speed pumps and several air-handling units. The plant serves classrooms and laboratories with strongly varying occupancy.

Problem

Electricity use rose by 14% over two years although floor area was unchanged. Operators suspected fouled heat exchangers, simultaneous heating and cooling, and poor scheduling. Existing meters recorded monthly electricity but not chilled-water flow or equipment-level power.

Analysis

The engineering team installed calibrated temperature sensors on chilled-water supply and return lines, ultrasonic flow meters, power meters on chillers and pumps, valve-position sensors and occupancy schedules. Data were sampled at five-minute intervals and synchronised with outdoor temperature. A data-quality screen rejected impossible temperatures, zero-flow records during known maintenance and periods of sensor communication loss.

The team calculated cooling output using Q̇ = ṁ c_p ∆T and plant COP using total plant power, including pumps and cooling-tower fans. A clustering model separated occupied, unoccupied and hot-weather operating modes. A rules-plus-model diagnostic then flagged periods in which a high cooling load coincided with open reheat valves and high heating-water demand.

Approach and solution

The intervention included corrected schedules, a minimum outdoor-air reset strategy, chilled-water temperature optimisation within humidity constraints, coil cleaning and repair of two leaking control valves. Operators retained authority over setpoints, and all automated commands were limited by hard safety bounds. A monthly review compared measured energy with a weather- and occupancy-adjusted baseline.

Outcome

After commissioning, the plant used less electricity during mild weather, and simultaneous heating and cooling was substantially reduced. The team did not attribute all improvement to AI: mechanical cleaning, valve repair and schedule correction were also necessary. The project demonstrated that analytics should reveal causes and support engineering work rather than serve as a substitute for it.

Learning points

1. Equipment-level metering is often necessary before advanced analytics can be trusted.
2. A physically meaningful KPI is more useful than a generic energy total.
3. Interventions should be verified against adjusted operating conditions.
4. Human review, safety limits and maintenance actions are essential parts of an AI project.
5. Sustainability benefits should include electricity, refrigerant, equipment-life and occupant-comfort effects.

9. Challenges, governance and best practices

Technical limitations

Models may fail after equipment replacement, control changes, unusual weather or production changes. Black-box predictions can be difficult to validate against conservation laws. Sensor drift can create apparent degradation, while correlated variables can make causal interpretation unreliable. Training data may also overrepresent normal operation and contain too few failures.

Cybersecurity and safety

Connected mechanical assets expand the attack surface. Secure design should include network segmentation, authenticated devices, least-privilege access, encrypted communications where appropriate, patch management, backups, event logging and a tested incident-response plan. Safety-critical functions should have independent protection and local fallback control. An AI system should not be permitted to disable a relief device, bypass combustion interlocks or override legally required protection.

Data and ethical governance

Occupancy and environmental data can reveal patterns of individual or group behaviour. Collection should be proportionate, access-controlled and retained only as long as necessary. Models should document data sources, assumptions, uncertainty, version history and responsible owners. Operators should be able to understand why a recommendation was made and challenge it when physical conditions differ from the training data.

Best-practice checklist

• Begin with an energy and process audit.
• Define the system boundary, KPI and decision owner.
• Calibrate sensors and record uncertainty.
• Use common timestamps, units and asset identifiers.
• Combine physical laws with data-driven models.
• Validate on unseen operating conditions.
• Use alarms that prioritise actionable faults.
• Test recommendations in simulation or supervised operation.
• Include safety, comfort, quality and emissions constraints.
• Verify savings and periodically recommission the system.

Contemporary developments

Digital twins increasingly combine first-principles models with live sensor data. Edge computing can reduce latency and keep control functions operating during network outages. Physics-informed machine learning can constrain predictions using conservation laws, although it does not remove the need for measurement and validation. Federated and privacy-preserving methods may allow organisations to learn from distributed assets without transferring sensitive raw data. Carbon-aware control, heat-pump integration, thermal storage and industrial waste-heat recovery are expanding the role of AI–IoT systems in decarbonisation.

These developments should be evaluated critically. More sensors and more complex models do not automatically create more value. The appropriate solution is the simplest reliable system that supports a defined engineering decision, remains safe under abnormal conditions and produces verifiable improvement.

Chapter Summary

Energy, thermal and sustainability applications of AI–IoT combine measurement, thermodynamics, control, data analytics and lifecycle decision-making. Energy and power must be distinguished, and efficiency must be defined with a clear system boundary. Control-volume balances, sensible heat equations, COP, boiler efficiency, turbine efficiency and pump power provide physically meaningful foundations for analytics.

IoT supplies contextual, time-stamped measurements; AI supports forecasting, anomaly detection, diagnosis, predictive maintenance and constrained optimisation. Baselines must account for weather, occupancy, production and operating mode. HVAC, boilers, turbines and fluid systems each require specialised KPIs and safety constraints. Emissions reduction should include direct, indirect and, where relevant, embodied impacts. Lifecycle assessment and economic analysis prevent narrow energy savings from producing unintended environmental or financial outcomes.

Successful projects depend on calibrated sensors, high-quality data, transparent models, human oversight, cybersecurity, safe fallback control and verified savings. The central professional skill is not merely selecting an algorithm, but connecting a trustworthy analytical result to a physically sound and sustainable engineering action.

Key Terms

Energy: Capacity to perform work or transfer heat, measured in joules.

Power: Rate of energy transfer, measured in watts.

First-law efficiency: Useful energy output divided by energy input.

Exergy: Maximum theoretical useful work obtainable as a system reaches environmental equilibrium.

Control volume: Defined region across whose boundary mass and energy may flow.

COP: Cooling or heating capacity divided by relevant input power.

HVAC: Heating, ventilation and air-conditioning system.

IoT: Network of connected physical devices that sense, communicate and sometimes act.

Digital twin: A dynamic digital representation linked to a physical asset or process.

Baseline: Estimated reference performance without a defined intervention.

Anomaly detection: Identification of behaviour that differs from expected operation.

Fault diagnosis: Determination of probable causes of abnormal behaviour.

Predictive maintenance: Maintenance supported by estimated future condition or failure risk.

Energy intensity: Energy use per unit of product, service, area or activity.

Carbon intensity: Greenhouse-gas emissions per unit of output or energy.

Lifecycle assessment: Evaluation of environmental impacts across a product or system life cycle.

Net present value: Present value of benefits minus costs over a defined period.

Embodied energy: Energy used to produce, transport, install and dispose of an asset.

Excess air: Combustion air supplied above the theoretical amount required for complete combustion.

System boundary: Explicit limit defining which inputs, outputs and impacts are included.

Review Questions

Short Answer Questions

1. Distinguish energy from power and state their SI units.
2. Why must a baseline be adjusted for weather or production conditions?
3. State two measurements useful for diagnosing HVAC inefficiency.
4. What is the purpose of a system boundary in sustainability analysis?
5. Why should AI control systems include local fallback and safety interlocks?

Long Answer Questions

1. Explain the five layers of an AI–IoT architecture for an industrial energy system.
2. Discuss how sensor quality, data context and model validation affect energy analytics.
3. Compare direct boiler-efficiency measurement with an indirect loss-based method.
4. Explain how variable-speed pumping can reduce energy use and identify its limitations.
5. Discuss why lifecycle assessment is necessary when evaluating an energy-saving technology.

Analytical / Application Questions

1. A water loop has a mass flow rate of 2 kg/s and is heated from 20 °C to 35 °C. Using c_p = 4.18 kJ/(kg·K), calculate the heat-transfer rate and state the assumptions.
2. A pump delivers water at 0.05 m³/s through a head of 30 m. If density is 1000 kg/m³ and overall efficiency is 0.72, calculate hydraulic power and input power using g = 9.81 m/s².
3. Design a monitoring and governance plan for a boiler whose fuel consumption has increased but production has remained constant. Identify sensors, KPIs, possible faults, AI methods, safety constraints and savings-verification steps.

Multiple Choice Questions with Answers

Question 1

Which quantity is the rate of energy transfer?

A. Energy

B. Power

C. Efficiency

D. Enthalpy

Correct Answer: B. Power

Question 2

A chiller provides 300 kW of cooling while consuming 100 kW of input power. What is its COP?

A. 0.33

B. 1.5

C. 3.0

D. 30

Correct Answer: C. 3.0

Question 3

Which measurement is most directly useful for calculating sensible heat transferred to a flowing fluid?

A. Mass flow rate and temperature difference

B. Sound level and humidity only

C. Ambient light and occupancy only

D. Motor colour and casing size

Correct Answer: A. Mass flow rate and temperature difference

Question 4

In the pump relation P_h = ρgQH, what does H represent?

A. Heat capacity

B. Humidity

C. Pump head

D. Hydrogen concentration

Correct Answer: C. Pump head

Question 5

Which AI task estimates future energy demand?

A. Forecasting

B. Calibration

C. Insulation

D. Condensation

Correct Answer: A. Forecasting

Question 6

Why is plant-level HVAC COP often lower than chiller-only COP?

A. Plant COP excludes all input power

B. Plant COP includes pumps, fans and cooling-tower power

C. Chiller COP is always measured in joules

D. Plant COP applies only to boilers

Correct Answer: B. Plant COP includes pumps, fans and cooling-tower power

Question 7

A baseline-adjusted savings calculation is necessary primarily because

A. energy meters cannot measure time

B. energy use depends on conditions such as weather and production

C. all equipment has identical efficiency

D. emissions factors never change

Correct Answer: B. energy use depends on conditions such as weather and production

Question 8

Which action best protects a connected combustion system?

A. Allowing unrestricted remote overrides

B. Removing local interlocks

C. Using authenticated access and independent safety protection

D. Training only on normal operation and ignoring faults

Correct Answer: C. Using authenticated access and independent safety protection

Question 9

If a pump has hydraulic power of 14.715 kW and overall efficiency of 0.70, its input power is approximately

A. 10.3 kW

B. 14.7 kW

C. 21.0 kW

D. 35.0 kW

Correct Answer: C. 21.0 kW

Question 10

Which statement about lifecycle assessment is correct?

A. It considers only the purchase price

B. It considers impacts across relevant stages from production through end-of-life

C. It excludes operational energy by definition

D. It is identical to simple payback

Correct Answer: B. It considers impacts across relevant stages from production through end-of-life