Capacity planning for a machine-tool cell starts with one figure: machine-hour capacity, calculated as the number of usable machines multiplied by the productive hours labor can actually apply to them, then compared against throughput time per part to size the gap between theoretical and demonstrated output [S1].
Most discrete-part machine shops working in mixed-model, engineer-to-order mode quote lead times from historical actual output rather than from this theoretical ceiling, which leaves the master schedule unvalidated and routinely inflates quoted due dates by 20-40% when a new part mix arrives [S1]. The fix is to layer Rough Cut Capacity Planning (RCCP) on top of the master schedule before MRP explodes the plan into detailed routings.
What RCCP Actually Checks Before MRP Runs
RCCP is a long-term, gross-capacity technique that verifies whether key and critical resources can support the master schedule, stated in hours per week per resource under routing-based RCCP or in production rate per week per line under rate-based RCCP [S2]. It deliberately ignores scheduled receipts and on-hand inventory, so its answer is the gross production requirement against installed capability, not the netted MRP load [S2].
The planning loop closes by negotiating either side of the equation: the master schedule can be re-dated or its quantity changed, and available capacity can be moved by adding shifts, overtime, subcontracted hours, or new machines [S2]. Resource groups in Oracle Bills of Material let planners restrict the RCCP review to only the bottlenecks, which is where the real signal lives in a pneumatic tool assembly cell or any multi-station job shop.
The Two Calculation Methods Behind the Number
The manual path requires two inputs: machine-hour capacity, which is the count of usable machines times the hours employees can run them, and throughput time, the elapsed time a single unit spends in the system from release to finished goods [S1]. The shorter that throughput time is relative to available machine-hours, the higher the design capacity ceiling, and the further demonstrated actual output can sit below that ceiling once changeovers, tool changes, and queue waits are netted out.
Capacity utilization rate, expressed as actual output divided by production capacity for the same period, is the metric that exposes whether the gap is structural (too few machines, wrong mix) or operational (poor OEE, excessive setup) [S1]. In power tool and rebar tool production lines that run long changeovers between SKUs, utilization rates above 80% almost always trigger queue explosion at the bottleneck station and force a capacity re-plan rather than a schedule re-plan.
Adding Electrical Energy as a Planning Resource
A 2015 TU Wien study argued that electrical energy demand of machine tools should be modeled as a planning resource alongside labor and machine hours inside MES, because in some manufacturing sectors electricity reaches up to 15% of total production cost and only 5-20% of the energy input to a machine tool is actually consumed by the cutting process itself [S4]. The implication is that load shifting, peak shaving, and renewable-supply matching become optimization variables in the same solver that currently balances hours and rates.
The same paper frames the industrial sector as roughly half of national electricity consumption in Austria (Statistic Austria 2009), giving planners leverage that smart-grid demand-response programs aimed at households cannot match, because production is already planned in IT systems that can ingest a kW profile per routing step [S4]. For a concrete tool foundry or a survey tool machining line running induction heat-treat, exposing the kVA draw per work order lets RCCP validate not just hours but also transformer and substation headroom.
Machine Learning for What-If Capacity Sizing
Red Hat's MLASP (Machine Learning Assisted System Performance and Capacity Planning) process replaces hand-tuned configuration sweeps with a trained model that predicts KPI outcomes from a configuration vector, then solves the inverse problem to find configurations within a given percentage deviation of a target throughput or latency [S5]. The process splits into three stages: automated load testing that generates the training set, model training by a data scientist, and model serving for what-if queries and target-seeking queries [S5].
The value for a machine-tool plant is that the same ML surrogate can front-run RCCP: instead of re-running the master schedule against hours per week per resource, planners ask the model for the bottleneck hours a new SKU mix will consume, and only commit to detailed MRP when the surrogate prediction lands inside the critical-resource envelope. This is a closer fit to tool die steel heat-treat scheduling, where each batch has a different soak curve and a different parasitic load on shared utilities.
Decision Matrix: Three Capacity-Planning Approaches
Choosing between the three approaches is a function of part-mix stability, data maturity, and planning horizon, and the table below lines them up against the criteria that actually drive the choice. [S6]
Manual machine-hour plus throughput-time math is the lowest-cost option and works for stable make-to-stock workflows with a simple product mix and reliable demand, but it ignores supply, labor-skill, and workstation changes between periods [S1]. RCCP over the master schedule is the middle option, giving hours-per-week visibility per resource or rate-per-week visibility per line, and it forces the negotiation between marketing and production before detailed MRP explodes the plan [S2]. ML-assisted capacity planning is the highest-cost option, justified when the configuration space is too large to brute-force, the load-test cycle has stabilized, and the OPEX savings from avoiding over-provisioned machines exceed the modeling investment [S5].
Where This Breaks: Limitations and Failure Modes
RCCP is explicitly a gross technique, so a clean RCCP pass does not guarantee that the detailed MRP will find enough capacity at every workcenter, only at the key and critical ones flagged in the resource group definition [S2]. Historical actual output is a reliable baseline only for stable make-to-stock flows producing a simple mix, and it carries no information about upcoming workstation, labor, or supply changes [S1]. Energy-aware planning adds data-model and interface work between MES and the building energy management system that most plants have not yet built, so the kW dimension currently sits outside most commercial RCCP implementations [S4]. ML-assisted planning inherits the bias of the load-test data set, and a model trained on a single SKU mix will mis-predict when a new geometry enters the cell [S5].
Standards and Sourcing Anchors
RCCP resource and routing definitions sit inside Oracle Bills of Material and Oracle Capacity, where resources are assigned to resource groups that scope the rough-cut review to key and critical workcenters only [S2]. The WebLogic capacity-planning checklist groups its drivers into application tuning, database limits, network load, SSL overhead, and client traffic mix, and notes that tunneling T3 over HTTP runs roughly 15% below native HTTP on the same hardware, a useful ratio when sizing the application server tier that fronts the MES [S3]. A 2018 Springer evaluation of capacity planning and practical scheduling algorithms in machine tool manufacturing systems remains a reference point for the algorithmic side, while the 2015 TU Wien paper remains the canonical reference for adding electrical energy demand as a first-class planning resource [S4][S6].
Two trackable signals for the next planning cycle: watch for MES vendors exposing a kW-per-routing-step field that RCCP can consume, and watch for CNC machine manufacturing quality standards updating to call out energy intensity as a procurement gate alongside geometric tolerance. For plants running mixed-model cells, the immediate next step is a one-week pilot that adds electrical load profile to the master schedule and re-runs routing-based RCCP against the bottleneck workcenter only.