Energy-Aware Production Planning: Connecting Factory Schedules with Energy Data

Why Total Energy Consumption Is Not Enough

A factory manager receives the monthly electricity bill, sees a figure in kilowatt-hours, and asks the obvious question: where did that energy actually go? The bill answers nothing useful. It aggregates compressors, CNC spindles, lighting, HVAC, idle conveyors, and a batch of changeover waste into a single number. You cannot schedule around a number that hides its own causes.

Total consumption is a lagging, undifferentiated indicator. It tells you the factory ran, not how it ran. Two identical production weeks can differ by fifteen percent in energy use simply because one week had more short batches, more machine idling between orders, or a hot summer afternoon when cooling loads spiked. Without decomposing energy by machine, process step, shift, and product, the production planner has no actionable information. Energy-aware production planning begins at this decomposition step: turning a single meter reading into a structured picture of where, when, and why kilowatt-hours are consumed.

For small and medium manufacturers, this does not require a utility-grade SCADA overhaul. In many cases, a handful of sub-meters on the main distribution panels, combined with existing PLC or CNC controller data, provides enough granularity to build meaningful energy profiles. The critical insight is that the value is not in the metering itself but in what the metering enables downstream: context, modelling, and better scheduling decisions.

Production Context and Energy Intensity

Energy consumption in a factory is not a fixed property of a machine. It depends on the production context in which that machine operates. A five-axis machining centre running a continuous roughing pass at full spindle speed draws a different power profile than the same machine executing a series of short finishing operations with frequent tool changes and coolant pump cycling. The energy per unit of output can vary by a factor of two or three depending on batch size, cycle time, idle gaps, and the product mix loaded onto that machine.

Several contextual factors shape energy intensity in manufacturing:

• Batch size and changeover frequency. Short batches multiply setup energy. Tool changes, coolant flushes, machine warm-up cycles, and program loading all consume power without producing saleable output.
• Machine loading. Many industrial machines draw a substantial fraction of their rated power even at low utilisation. A compressor at 30 percent load may consume 60 to 70 percent of its full-load energy. Scheduling that leaves machines half-idle inflates per-unit energy cost.
• Product mix and process sequence. A job shop that interleaves aluminium and steel workpieces on the same machine incurs different cutting forces, coolant demands, and spindle loads than one that groups similar materials together.
• Ambient and seasonal conditions. Compressed-air systems, paint booths, drying ovens, and HVAC loads respond to outside temperature and humidity, adding a non-production energy layer that complicates attribution.
• Shift structure and idle periods. Lighting, conveyors, and standby electronics run during breaks and between shifts unless explicitly managed.

Recognising these dependencies is the first analytical step. The production planner does not need to model every thermodynamic interaction, but they do need to understand that energy per part is a variable, not a constant, and that scheduling choices shift that variable in predictable directions.

From Machine Data to Scheduling Decisions

Once energy use is decomposed and contextualised, the question becomes: how does this information actually change the schedule? The bridge between measurement and decision is a simple energy-intensity model attached to each process step or product family. In practice, this looks like a table or a lightweight regression: for product X on machine Y, the expected energy per hour (or per batch) is approximately Z, with known sensitivity to batch size and idle gaps.

Building such a model does not require a data-science team. A production engineer who logs machine run-time, spindle load, and compressed-air pressure over a few weeks can derive per-product energy coefficients with basic spreadsheet analysis or a short Python script. The model is approximate, and that is fine. Scheduling decisions are robust to moderate model error; what matters is the relative ranking of options, not the absolute kilowatt-hour prediction to two decimal places.

With energy coefficients in hand, the planner can evaluate scheduling alternatives:

Time-shifting. If the factory has access to time-of-use tariffs or on-site renewable generation, energy-intensive operations (large compressors, induction heating, heavy milling) can be preferentially scheduled during low-cost or high-renewable windows. This does not require real-time automation; a weekly planning adjustment based on a forecast is often sufficient.

Batching and sequencing. Grouping similar products reduces changeover energy. Scheduling a sequence of aluminium parts before a sequence of steel parts, rather than interleaving them, cuts coolant and tool-change waste. The planner sees this trade-off when energy-per-changeover is quantified.

Idle management. Knowing that a machine draws 15 kW in standby versus 3 kW when truly off makes it worthwhile to schedule longer gaps between orders as explicit shutdowns rather than idle periods. This is a policy decision informed by data.

Capacity allocation. If two machines can perform the same operation but one is significantly more energy-efficient at the required tolerance, the scheduler can weight energy cost alongside throughput and setup time.

The key point is that these are planning-layer decisions, not machine-control-layer decisions. The energy data feeds the schedule; the schedule then dictates machine behaviour. This separation keeps the system understandable and avoids the temptation to automate every adjustment.

Forecasting and Decision-Support Layers

Reactive energy monitoring tells you what happened. Predictive energy modelling tells you what will happen if you schedule this way. The next layer of maturity is a decision-support tool that evaluates candidate schedules against a forecast energy profile and presents the planner with trade-offs: this sequence saves 8 percent energy but delays delivery by half a day; this alternative uses peak-hour power but avoids a costly changeover.

Forecasting in this context does not mean a neural network predicting grid load six weeks out. For most SMEs, a rolling 24-to-72-hour energy forecast built from historical machine profiles, the upcoming production plan, and a simple tariff or generation schedule is adequate. The model inputs are: planned orders, machine assignments, expected cycle times, and the energy coefficients derived earlier. The output is a projected hourly or shift-level energy demand curve. Comparing that curve against tariff windows or renewable generation forecasts gives the planner actionable adjustment options.

European R&D projects have explored exactly this architecture. The EnerMan project developed an energy-sustainability management framework that combines factory floor data, process and equipment models, prediction, and decision support into a coherent workflow. The framework is designed so that the prediction layer informs the scheduling layer without necessarily executing changes automatically. Similarly, DENiM (Digital Energy Management) built digital services integrating IoT sensor data, analytics, digital-twin simulation, energy modelling, and selective automation for manufacturing energy management. The emphasis in DENiM was on providing the planner with a digital twin of the energy system so that scheduling scenarios could be evaluated before commitment.

The ECOFACT project took a broader integration view, linking production data, energy data, and environmental information through sensors, analytics, digital twins, and dynamic life-cycle models. Its architecture demonstrates that energy-aware scheduling is not an isolated function but a node in a larger information graph connecting product design, process planning, and environmental reporting.

The EU’s DT-FOF-09-2020 programme guidance frames this work within a wider industrial context, linking ICT, digital twins, big data, IoT, and AI with near-real-time production-system energy management. The guidance explicitly acknowledges that the appropriate level of integration varies by factory size, product complexity, and energy cost significance, and that not every enterprise needs a fully closed-loop system.

For an SME, the practical implication is this: a decision-support dashboard that shows projected energy use per schedule variant, flags tariff-sensitive operations, and highlights idle-energy waste is often the highest-value, lowest-complexity intervention. It changes behaviour without replacing the planner.

When Automation Is Justified

Closed-loop energy-aware scheduling, where the system automatically reschedules or adjusts machine parameters in response to real-time energy signals, is the most technically impressive option and the least appropriate for most small manufacturers. It is justified when several conditions converge:

• Energy cost represents a significant fraction of product cost (for example, energy-intensive processes like glass melting, aluminium smelting, or large-scale drying).
• The production system has genuine flexibility: multiple machines, interchangeable products, low changeover penalties.
• Energy prices or availability fluctuate rapidly enough that human planning cycles (daily or weekly) cannot capture the savings.
• The factory already operates a mature MES or APS system that can accept schedule modifications programmatically.
• The organisation has the technical capacity to maintain and validate the optimisation logic.

When these conditions are absent, automation adds fragility, maintenance burden, and risk of suboptimal behaviour when the model drifts from reality. A human planner who understands the energy context and receives a clear dashboard is more robust than an unattended optimiser that silently mis-schedules when a sensor fails or a product mix shifts.

Even in larger operations, a graduated approach is common: start with monitoring, add modelling, introduce decision support, and only then consider selective automation for well-understood, high-frequency decisions (for example, automatically switching a compressor bank between load stages). The European projects cited above all include automation as an optional layer, not a mandatory starting point.

A Hypothetical SME Production Scenario

Consider a mid-size CNC machining shop in Upper Austria with twelve machines, three shifts, and a product mix of aerospace brackets, automotive housings, and custom fixtures. Annual electricity cost is in the range of 180,000 euros, representing roughly 4 to 5 percent of revenue. The shop already runs a basic ERP for order management and has PLC data logging on most machines.

The energy-aware planning initiative proceeds in stages over about nine months:

Stage 1 – Measurement (months 1–3). Sub-meters are installed on the main compressed-air system, the paint/drying booth, and three high-power machining centres. Existing PLC data (spindle speed, coolant pump status, axis load) is exported to a historian. Six weeks of data collection establishes baseline energy profiles per machine and per product family. The shop discovers that compressed air accounts for 22 percent of total electricity and that two machines idle at 8 kW between operations during the night shift.

Stage 2 – Modelling (months 4–5). A production engineer builds simple energy-per-batch coefficients for the top twenty product families, using the historian data and known cycle times. The model is a spreadsheet lookup with a few regression adjustments for batch size. Accuracy is within 10 to 15 percent, which is sufficient for scheduling comparisons.

Stage 3 – Decision support (months 6–8). A lightweight dashboard (built on existing MES data plus the energy coefficients) shows the planner, for each candidate weekly schedule, the projected hourly energy demand, the cost under the time-of-use tariff, and flagged opportunities such as: moving roughing operations on three parts to the 22:00–06:00 off-peak window, or batching all aluminium parts on Monday to reduce coolant-flush changeovers. The planner reviews the dashboard each Friday and adjusts the week’s schedule manually.

Stage 4 – Selective policy changes (month 9+). Based on the data, the shop implements a shutdown policy for compressed-air dryers during lunch and night breaks, and adds a standby-power threshold alarm on the machining centres. No closed-loop automation is introduced. The estimated annual saving from stages 1 through 4 is in the range of 8 to 12 percent of electricity cost, with the bulk coming from the idle-management and batching decisions rather than from time-of-use shifting.

This scenario is illustrative, not a measured case study. It reflects the kind of incremental, low-complexity path that European R&D frameworks like EnerMan and DENiM describe as appropriate for SMEs: measurement enables context, context enables modelling, modelling enables better human decisions, and automation is reserved for the narrow cases where it clearly pays for itself.

Evidence from European Manufacturing R&D

The projects referenced earlier are not isolated pilots. They represent a coordinated European research effort under Horizon 2020 and Horizon Europe to make manufacturing energy management a first-class concern in production planning, not an afterthought in the facilities department.

EnerMan (project 958478) developed a framework that explicitly connects factory data acquisition, process and equipment energy models, short- and medium-term prediction, and decision support for production planning. Its architecture separates the data layer, the model layer, and the planning layer, allowing factories to adopt the framework incrementally. The project emphasises that the goal is not to replace planners but to give them energy-awareness as an additional dimension in their existing decision space.

DENiM (project 958339) focused on digital services for energy management in manufacturing, integrating IoT sensor networks, analytics pipelines, digital-twin simulation, and energy-modelling tools. The digital-twin component allows planners to test scheduling scenarios against a virtual energy model before committing to the shop floor. Automation is included as an optional service layer for specific use cases like demand-response participation or peak-shaving, but the core value proposition is the modelling and simulation capability.

ECOFACT (project 958373) broadened the scope to include environmental and life-cycle information alongside energy data. By integrating sensors, analytics, digital twins, and dynamic life-cycle assessment models, ECOFACT demonstrates that energy-aware scheduling can simultaneously reduce carbon footprint and material waste, and that the information architecture supporting one supports the other. This integration matters for SMEs facing customer or regulatory pressure to report environmental performance, because the same data infrastructure serves both energy planning and sustainability reporting.

The DT-FOF-09-2020 programme guidance provides the policy and strategic context, framing energy-efficient manufacturing as a priority area where ICT, digital twins, big data, IoT, and AI converge to enable near-real-time production-system energy management. The guidance acknowledges heterogeneity across sectors and factory sizes, and does not prescribe a single architecture. Instead, it encourages modular, scalable approaches that match the enterprise’s complexity and energy-cost sensitivity.

Taken together, these projects and guidance documents confirm a consistent finding: the highest-impact interventions for most SMEs are at the measurement, modelling, and decision-support layers. Full closed-loop optimisation is a specialised tool for a subset of high-energy-cost, high-flexibility operations. The research community is not advocating a one-size-fits-all automation mandate; it is advocating for information flow and analytical capability that make energy a visible variable in the planner’s equation.

How Cademix GDTL Can Support Integration and Modelling

For an SME that has decided energy-aware planning is worth pursuing but lacks the in-house data engineering or modelling expertise, external support can accelerate the path from measurement to decision. Cademix Institute of Technology, as an interdisciplinary IT, AI, research, consulting, and adult-education institute based in Austria, can contribute across several of the workflow stages described above.

On the data and integration side, GDTL’s expertise in software development, data analysis, and digital workflows covers the practical task of connecting sub-meter data, PLC historian exports, and ERP production-order records into a unified data layer. This is often the most tedious and underappreciated step: getting the data to talk to each other in a consistent format with correct timestamps and machine-to-order mappings. Interdisciplinary coordination between IT specialists and production engineers is where GDTL’s model is designed to operate.

On the modelling and analytics side, GDTL’s work in AI, simulation, and data analysis supports the construction of energy-intensity models, the design of lightweight forecasting pipelines, and the development of decision-support dashboards. The modelling does not need to be a black-box neural network; in many SME contexts, a transparent regression or lookup model is preferable because the planner needs to understand and trust the numbers. GDTL’s approach emphasises interpretable models and iterative refinement rather than over-engineered solutions.

On the training and knowledge-transfer side, GDTL’s continuing-education and employability focus means that the production planners, shop-floor supervisors, and maintenance staff who will interact with the energy dashboard can be trained to interpret the data, understand the model assumptions, and make informed scheduling adjustments. This is a critical success factor: a dashboard that nobody reads is worse than no dashboard at all, because it creates the illusion of management without the substance of action.

On the documentation and project-collaboration side, technical documentation of the measurement architecture, data flows, model logic, and operational procedures ensures that the system remains maintainable and transferable. If the SME later participates in a European R&D collaboration or a supply-chain sustainability audit, the documented data pipeline and modelling methodology provide the evidence trail that such engagements require.

It is important to be precise about what GDTL is and is not. It is not a certification body, not a regulatory authority, and not a vendor that guarantees a specific percentage of energy saving. What it offers is interdisciplinary technical capability: the ability to design, build, and support the integration and modelling infrastructure that makes energy-aware planning feasible, and to train the people who will use it. The actual energy savings depend on the factory’s production characteristics, tariff structure, and the quality of the decisions the planners make with the tools provided.

Conclusion

Energy-aware production planning is a spectrum, not a binary switch. At one end, a factory manager reads the monthly bill and shrugs. At the other end, a fully automated closed-loop optimiser adjusts machine parameters and reschedules orders in real time against grid signals. Between those poles lies the practical territory where most manufacturing SMEs operate and where most of the achievable value is found: decompose your energy data, understand how production context drives energy intensity, build simple models that connect product and process to kilowatt-hours, and feed those models into a decision-support layer that makes energy a visible cost in the scheduling conversation.

The European R&D landscape, from EnerMan to DENiM to ECOFACT, validates this incremental architecture and provides reference implementations. The DT-FOF-09-2020 programme guidance confirms that the appropriate depth of integration scales with the enterprise’s needs. The critical editorial point for any SME reader is this: you do not need a digital twin of your entire factory to make your next production schedule three percent more energy-efficient. You need sub-meters, a spreadsheet, a dashboard, and a planner who asks “what does this sequence cost in power?” before finalising the week.

Start with measurement. Add context. Build a model you can explain. Put it in front of the people who schedule. Automate only the decisions that are repetitive, well-understood, and high-frequency. And treat energy-awareness not as a separate project but as an additional dimension in the production-planning discipline you already practice. That is the workflow logic that turns a kilowatt-hour number on a bill into a scheduling lever you can actually pull.