Simulation for Factory Energy Optimization: A Case-Based Walkthrough of Decisions, Trade-offs, and Transferable Lessons

When Energy Bills Become an Engineering Problem

European manufacturing sits at an uncomfortable intersection. Energy prices remain volatile after the disruptions of the early 2020s, sustainability reporting obligations are tightening across the EU, and competitive pressure means that a factory cannot simply pass rising costs to customers. For a mid-sized producer of, say, precision components or packaged goods, electricity and thermal energy now represent a line item large enough to justify the same analytical rigour once reserved for production scheduling or quality control. The question shifts from “how do we reduce energy use?” to “where exactly is energy wasted, what drives the waste, and what would actually change if we altered this process parameter?”

That second question is where simulation for factory energy optimization enters the picture. Not as a buzzword, not as a dashboard with a green leaf icon, but as a modelling discipline: building a computational representation of a factory’s energy flows, process interactions, and temporal behaviour so that decisions can be tested before they are implemented on the shop floor. This article follows a composite case to trace how that discipline works in practice, where it delivers, where it disappoints, and what a plant manager or technical lead should carry away.

Setting the Scene: A Composite Case

The scenario below is a deliberately constructed composite, drawing on common patterns observed across European small and medium-sized manufacturers. It is not a named customer, a published project report, or a measured case study. Treat it as a realistic teaching scenario designed to expose the decision logic that simulation-based energy work actually involves.

Imagine a factory in the DACH region producing roughly 12,000 tonnes of coated metal sheet per year. The site has a 2.1 MW electrical connection, a natural-gas boiler serving a drying line and space heating, compressed-air generation, and a recently installed 400 kWp rooftop photovoltaic array. Energy costs have climbed sharply over the past three years. Management has committed to a 20 percent reduction in specific energy consumption (kWh per tonne of output) within five years, partly to control costs, partly to satisfy a major automotive customer’s supply-chain sustainability questionnaire. The plant has an energy manager, a maintenance team, and a production scheduler. It does not have a dedicated simulation engineer, a digital-twin platform, or a coherent energy data model. It has, however, sub-metering on the main production lines, hourly gas and electricity readings from the utility, and a SCADA system that logs machine states.

The first decision the team faces is not “which software to buy.” It is a more fundamental question: what kind of analysis will actually inform the right interventions? A spreadsheet that aggregates monthly kWh and divides by tonnes of output tells you the trend. It does not tell you why a 4 percent spike appeared in March, whether the drying oven’s cycling pattern is costing more than a setpoint change would save, or whether shifting a batch of coating work from the second shift to the third would reduce peak demand charges without disrupting delivery. Those are questions about system behaviour over time, about interactions between processes, about counterfactuals. They require simulation.

Why Energy Simulation Is Not Just a Spreadsheet with More Rows

It is worth stating the distinction clearly, because in practice many factories conflate energy monitoring with energy modelling. Monitoring tells you what happened. Simulation lets you ask what would happen. The difference is structural.

A spreadsheet or a simple energy dashboard records historical consumption, computes averages, flags anomalies, and produces reports for compliance. Useful, necessary, but reactive. A simulation model, by contrast, encodes the causal relationships between process parameters, equipment behaviour, ambient conditions, production schedules, and energy flows. It can be driven with hypothetical inputs: what if the compressor operates at a different pressure band, what if the drying line runs at 78 °C instead of 85 °C, what if the PV array’s output profile changes after a panel replacement, what if a new batch arrives at 14:00 instead of 06:00.

For factory energy optimization specifically, three simulation families tend to matter. The first is process-level discrete-event or continuous simulation, where individual machines, buffers, and scheduling logic are modelled to capture how production sequencing affects energy draw over hours and days. The second is building and HVAC simulation, typically using tools in the EnergyPlus or TRNSYS family, which model thermal loads, ventilation, and the interaction between process heat demand and space conditioning. The third is network and distribution modelling, which looks at how loads interact on the electrical infrastructure, how demand charges respond to peak-shaving strategies, and how distributed generation like the rooftop PV changes the picture. In a real factory, these three layers are coupled: a scheduling decision changes the thermal load, which changes the boiler cycling, which changes the gas demand, which changes the electrical load profile, which changes the demand charge. A simulation that models only one layer will produce answers that are locally correct and globally misleading.

The Simulation Workflow in Practice

Returning to the composite case, the team’s first practical step was not modelling. It was data archaeology. The SCADA logs contained machine states but lacked timestamps aligned with the sub-metering data. The energy manager’s spreadsheets recorded monthly totals but not hourly profiles. The boiler’s control system stored cycling data, but only for the past fourteen months, and the format was proprietary. The compressed-air system had a pressure sensor but no flow meter, so air demand had to be inferred from compressor run-hours and a nameplate capacity assumption.

This is the unglamorous part of simulation for factory energy optimization that project proposals often understate. The model is only as good as the data that calibrates it, and in most existing factories the data is incomplete, inconsistent in time resolution, or locked in incompatible formats. The team spent roughly six weeks on data reconciliation before writing a single line of model code. They installed two additional flow meters on the compressed-air line and a thermal sensor on the drying oven exhaust, which were modest investments but closed the most critical measurement gaps.

With the data foundation in place, the modelling proceeded in layers. The production scheduling layer was built as a discrete-event model capturing the coating line, drying oven, cutting stations, and the buffer between them. Each machine had an energy signature: a base load when idle, a load profile during operation, and a ramp-up or ramp-down transient. These signatures were calibrated against the sub-metering data for representative production runs. The thermal layer modelled the drying oven as a lumped-parameter system with a heat loss coefficient, an air-flow term, and a dependence on ambient temperature. The electrical distribution layer tracked the aggregate load, the PV generation profile, and the utility’s time-of-use tariff structure.

Validation was iterative. The team ran the model against three months of historical data where the production schedule was known, compared simulated hourly consumption to measured hourly consumption, and examined residuals. A mean absolute percentage error below roughly 10 percent on the aggregate electrical load was considered acceptable for the purpose of comparative scenario analysis. The thermal model was less accurate at the hourly level because the oven’s cycling behaviour was stochastic, but it tracked the weekly energy balance within a reasonable band. The team did not aim for a digital twin in the strict sense; they aimed for a model good enough to rank intervention options, which is a different and often more achievable standard.

Decisions Under Uncertainty

Once the model was operational, the factory faced a set of candidate interventions, each with a cost, an implementation timeline, and an energy impact that could be estimated but not guaranteed. The simulation did not eliminate uncertainty; it structured it.

One candidate was a variable-speed drive retrofit on the two largest compressed-air units, replacing fixed-speed operation with pressure-band control. The model showed that this could reduce compressed-air energy by roughly 15 to 20 percent under the observed production mix, but the benefit depended heavily on whether the air demand profile was spiky or smooth. The missing flow-meter data meant the spikiness was estimated, not measured, and the team flagged this as a sensitivity risk. They installed the flow meters and re-ran the model before committing the capital expenditure. The second candidate was a change in drying-oven setpoint from 85 °C to 78 °C, which the thermal model suggested would save gas but might increase cycle time by 12 percent. The scheduling model then showed that the longer cycle time would create a bottleneck at the cutting station during peak weeks, requiring overtime or a buffer-stock strategy. The net energy saving was smaller than the oven-only estimate suggested, and the operational disruption was non-trivial. The third candidate was a shift in the coating-line batch schedule to align more production with peak PV generation hours. The electrical distribution model showed a modest reduction in grid draw during summer weekdays but almost no benefit in winter, when PV output was low and the tariff structure was flat. The intervention was worth doing but should not be oversold.

These are the kinds of trade-off conversations that simulation makes concrete. Without the model, the team would have been arguing from intuition and partial data. With the model, they could rank options, quantify the uncertainty in each ranking, and identify which additional measurements would most improve confidence. That last point matters: a good simulation workflow does not just answer questions, it tells you which questions you cannot yet answer and what it would cost to answer them.

What the Simulation Revealed — and What It Didn’t

After several months of scenario analysis, the composite factory identified a portfolio of interventions with a combined estimated reduction in specific energy consumption in the range of 12 to 18 percent, depending on production mix and weather. That fell short of the 20 percent management target. The gap was not a modelling failure; it reflected the genuine limits of what process-level changes could achieve without a deeper re-engineering of the coating chemistry or a heat-recovery system that had not been in the original option set.

More instructively, the simulation surfaced patterns the team had not anticipated. The compressed-air system’s energy draw was more sensitive to ambient temperature than expected, because the compressor’s volumetric efficiency dropped in summer heat. The drying oven’s cycling pattern interacted with the boiler’s minimum-load constraint in a way that created a small but persistent energy penalty: the boiler could not modulate below 40 percent, so during low-demand periods it cycled on and off, wasting gas. These were not dramatic findings, but they were the kind of second-order effects that a spreadsheet-based energy review would miss entirely. They also generated quick wins that required no capital expenditure: adjusting the boiler’s minimum-load setpoint and adding a small thermal buffer.

What the simulation did not reveal was the human and organisational layer. The model assumed that operators would follow the new scheduling logic consistently. In practice, shift-change handovers, manual overrides, and the informal “we always run the oven this way” culture introduced deviations that the model could not predict. This is a well-known limitation of factory energy simulation: it models the designed behaviour of the system, not the enacted behaviour. The gap between the two is a management and training problem, not a modelling problem, and it is worth acknowledging explicitly rather than pretending the model captures everything.

From Model to Measured Savings: The Implementation Gap

Simulation for factory energy optimization produces ranked recommendations. It does not produce savings. The distance between a modelled 15 percent reduction and a measured 15 percent reduction is filled with procurement, installation, commissioning, operator training, maintenance access, production planning adjustments, and the inevitable compromises that arise when a retrofit has to be done without stopping the line for three weeks.

In the composite case, the variable-speed drive retrofit delivered close to the modelled savings after a two-month commissioning period, because the compressor control logic was straightforward and the measurement infrastructure was already in place. The drying-oven setpoint change, by contrast, delivered only about half the modelled saving because operators reverted to the higher temperature during rush orders, and the thermal model had not captured the frequency of those reverts. The scheduling shift aligned with PV generation worked as predicted in summer but was abandoned in winter when the energy manager judged the operational complexity not worth the marginal benefit.

This pattern is common and worth internalising before a factory invests in simulation: the model’s accuracy is necessary but not sufficient. The implementation environment introduces its own dynamics. A realistic energy-simulation programme includes a feedback loop where measured post-implementation data is fed back into the model to recalibrate and to build institutional trust in the approach. Without that loop, the first successful simulation project becomes an anecdote rather than a repeatable capability.

Failure Modes and When Simulation Is the Wrong Tool

Not every energy question in a factory benefits from simulation, and it is honest to name the conditions under which it does not. If the primary energy driver is a single large motor or compressor whose load profile is well understood and whose control strategy is already near-optimal, a targeted engineering review will be faster and cheaper than building a model. If the factory’s energy problem is fundamentally one of measurement — the sub-metering does not exist, the data quality is poor, and nobody can say where the energy goes — then the first investment should be in instrumentation, not simulation. A model calibrated on garbage produces confident garbage.

Simulation also struggles when the intervention is highly non-linear or involves new process chemistry that has no existing energy signature to calibrate against. If the coating line is being replaced with an entirely different technology, the model cannot predict its behaviour; it can only help with the scheduling and thermal integration of the new unit once its characteristics are known. And for very small factories with a single production line and a simple tariff structure, the overhead of building and maintaining a simulation model may exceed the energy savings it identifies. The threshold is not absolute, but it is real: the complexity of the system has to justify the complexity of the model.

Another failure mode is over-precision. A model that reports energy savings to three decimal places invites a false confidence that the underlying data does not support. The composite case’s 12-to-18-percent range was more honest than a single-point estimate, and the team learned to present ranges with explicit sensitivity notes. Managers who demand a precise number often get a precise number that is wrong, which is worse than an honest range that is right.

Transferability: What Other Factories Should Take Away

The composite case is not a universal recipe, and no single factory’s energy journey maps cleanly onto another’s. But several structural lessons transfer across contexts. First, the data foundation determines the ceiling of what simulation can deliver. Before investing in modelling software or consulting, a factory should audit its measurement infrastructure and close the most critical gaps. Two flow meters and a thermal sensor, as in the composite case, can change the quality of every subsequent analysis. Second, model the system in layers and respect the couplings. A thermal model that ignores the scheduling model, or an electrical model that ignores the thermal model, will produce internally consistent but externally misleading answers. Third, treat simulation as a decision-support tool with an explicit uncertainty budget, not as a prediction machine. The value is in ranking options, identifying sensitivities, and revealing second-order effects, not in producing a guaranteed savings figure. Fourth, build a feedback loop. Post-implementation measurement recalibrates the model and turns a one-off project into a persistent capability. Fifth, pair the technical model with an operational change plan. The model may say that a setpoint change saves energy; the plan has to address why operators resist the change and how to make it stick.

For European SMEs specifically, the regulatory and market context adds a layer of urgency that is not purely technical. Energy audits required under frameworks such as the EU Energy Efficiency Directive, sustainability reporting under the Corporate Sustainability Reporting Directive, and customer supply-chain questionnaires all create a demand for credible energy data and credible improvement pathways. Simulation contributes to both: it produces the analytical backbone for the audit, and it generates the defensible improvement narrative for the report. But it should be positioned as an engineering tool serving those compliance and commercial needs, not as a compliance product in itself.

The Interdisciplinary Dimension

Factory energy simulation sits at the intersection of several competencies that rarely reside in a single person or a single department. The thermal engineer understands the drying oven. The electrical engineer understands the distribution network and the demand charge. The production planner understands the scheduling constraints. The data analyst understands the SCADA logs and the measurement gaps. The sustainability lead understands the reporting framework. The software engineer builds and maintains the model. In a small or mid-sized factory, these roles are often held by the same person wearing different hats, or by a very small team that cannot cover all of them simultaneously.

This is where interdisciplinary coordination becomes a practical necessity rather than an abstract ideal. A simulation for factory energy optimization that is built without the production planner’s input will miss scheduling constraints. One that is built without the thermal engineer’s input will model the oven incorrectly. One that is built without the data analyst’s input will be calibrated on unreliable measurements. The model is only as robust as the breadth of expertise that shaped its assumptions.

At Cademix Institute of Technology, this kind of cross-domain work is part of the operational reality. The institute brings together expertise in digital strategy, software development, data analysis, AI and automation, CAD and 3D modelling, simulation, technical documentation, and continuing education within a single interdisciplinary structure. For a factory that needs to build or refine an energy-simulation capability but does not have all the competencies in-house, that kind of coordinated expertise can reduce the friction of assembling a project team from separate vendors. Cademix does not provide regulatory certification or licensed engineering sign-off, and it does not guarantee specific savings outcomes. What it can contribute is the technical and organisational scaffolding: modelling, data engineering, simulation development, training of in-house staff so that the capability persists after the project, and the coordination of specialists who need to work together on a coupled problem. For R&D teams exploring novel process configurations or for manufacturers evaluating whether a digital-twin approach to energy management is viable for their specific production mix, that kind of structured, interdisciplinary support is where the work becomes tractable rather than aspirational.

Conclusion: Reusable Principles, Not Universal Claims

Simulation for factory energy optimization is not a turnkey solution and not a marketing category. It is a modelling discipline that, applied carefully, changes the quality of energy decisions in a manufacturing environment. The composite case traced here illustrates the workflow, the trade-offs, the failure modes, and the organisational realities that any factory will encounter when it moves from “we monitor our energy” to “we model our energy system to test interventions before we implement them.”

The reusable principles are these: invest in measurement before modelling; model the couplings, not just the components; present uncertainty honestly; close the loop between modelled predictions and measured outcomes; and treat the simulation as one input into a decision process that also includes operational feasibility, capital constraints, and human behaviour. A factory that internalises those principles will get more from simulation than a factory that treats it as a software purchase. And a factory that skips those principles will build an elegant model that nobody trusts and nobody uses.

None of this is specific to one industry, one country, or one production technology. The structure of the problem — coupled energy flows, incomplete data, operational variability, organisational resistance, and a management target expressed as a percentage — recurs across European manufacturing. The answers are always context-specific. The method of asking the questions well is not. That is what simulation for factory energy optimization is ultimately for: not to tell you what to do, but to make the trade-offs visible enough that you can decide what to do with open eyes.