When a single prototype line transitions into serial production across three or four plants, the tolerance strategy that worked at launch often becomes the most expensive bottleneck in the supply chain. A engineering team at the lead plant builds the PPAP package, sets the dimensional limits, and validates the process. The problem is that they base those limits on the specific capabilities of the launch equipment. They then push those same constraints to satellite facilities running older machinery, different tooling, and entirely different process variation.
The result is predictable and costly. The secondary sites cannot hit the inherited tolerances, generating scrap, concession requests, and 8D reports that flood the quality system. The default reaction from the engineering team is to tighten the print further, attempting to force compliance through sheer dimensional constraint. This defensive response ignores the actual mathematical behaviour of the assembly. It actively punishes the satellite plants for variation they cannot eliminate, while doing nothing to prove the final product function to the customer.
Statistical tolerancing, specifically the root-sum-square method, is the only practical way to manage dimensional chains across a multi-site manufacturing network. It decouples the functional assembly requirement from the specific capability of any individual machine. Across two decades in automotive and aerospace, I have seen organisations use RSS calculations to standardise drawings across global facilities. This approach allows each site to utilise its inherent process variation profile while mathematically guaranteeing that the final assembly meets specification.
The Network Capability Problem
Standard worst-case linear stack-up analysis demands that every dimension in the chain hits its extreme limit simultaneously. A central engineering team evaluating a five-part assembly using worst-case math sets the individual component limits at ±0.05 mm to achieve a ±0.25 mm assembly gap. This calculation requires precision grinding at every facility, instantly excluding older high-volume machining centres that easily hold ±0.10 mm. The supply chain inherits a massive capability restriction purely because the calculation method ignores statistical reality.
Root-sum-square calculations model actual variation, proving that independent component deviations compensate for one another. Applying the RSS method to those same five components, each toleranced at ±0.10 mm, produces an assembly variation of just ±0.224 mm. This mathematical proof sits comfortably inside the functional requirement. Every plant in the network can now use standard turning and milling, eliminating the precision grinding mandate and unlocking capacity that the worst-case methodology falsely restricted.
The scaling challenge is not the math; it is the data collection. To validate the RSS calculation, the central quality team must gather SPC data from every site producing the components. If Plant A runs a rigid CNC cell and Plant B runs a legacy single-spindle machine, their combined process data must demonstrate statistical control and independence. The evidence chain proves the math, but the network must support it with disciplined measurement systems analysis and standardised data reporting across all facilities.
Standardising the Evidence Across Facilities
Customers evaluating a multi-site PPAP submission want to see consistent process capability, not a fragmented collection of localised workarounds. When a supplier presents a statistical tolerance strategy, the customer expects the central quality team to govern the inputs. If the RSS calculation relies on data from the lead plant but the satellite facility produces entirely different variation curves, the entire submission fails. The gap between the mathematical proof and the shop-floor reality collapses the commercial argument.
Building a network-wide RSS strategy requires standardising the measurement inputs. Every facility must measure the critical-to-quality dimensions using the same gauge R&R methodology. If one plant uses a CMM and another uses a functional gauge, the central engineering team cannot aggregate the data into a single root-sum-square model. The standard dictates that the measurement system variation must not exceed 10% of the total tolerance. If the secondary site lacks the metrology capability, they must upgrade their quality lab before they can participate in the statistical tolerancing programme.
The customer's source inspection team will audit the weakest link. They will pull the SPC charts from the facility with the oldest equipment and demand proof of capability. A robust network strategy anticipates this. The central team maps the process capability index of every site, documenting the Cpk of the critical dimensions. The RSS calculation incorporates the lowest capable process into the statistical model. If the worst-performing site still meets the assembly requirement statistically, the customer gains absolute confidence in the entire manufacturing network.
Single-Site vs Multi-Site Tolerancing Strategy
Single-site launch approach
- Worst-case linear stack-up based on lead plant
- Tolerances tightened to cover hypothetical extremes
- Individual component limits drive unnecessary grinding
- Data collected from one controlled manufacturing source
Network-scale RSS approach
- Root-sum-square based on combined plant data
- Tolerances reflect true statistical assembly variation
- Standard machining permitted if assembly fits
- Data standardised across all producing facilities
When Satellite Sites Break the Assumptions
Root-sum-square math assumes independence, normal distribution, and statistical control. These assumptions are fragile in a multi-site environment. Consider a scenario where the lead plant machines a housing on a five-axis CNC centre, ensuring dimensional independence. The satellite facility machines the same housing on a three-axis machine using a custom fixture, introducing mechanical dependencies between the bore and the face. The RSS calculation collapses because the satellite process violates the independence requirement.
Process drift destroys the statistical model even faster than mechanical dependency. A satellite plant running worn tooling produces a skewed, non-normal distribution. The central quality team aggregates this data into the network report, completely invalidating the RSS prediction. Presenting this flawed calculation to the customer during a PPAP review or source inspection constitutes a severe quality escape. The customer's quality auditor will immediately spot the non-normal distribution, reject the submission, and potentially halt the production ramp-up.
Managing this risk requires the central engineering team to establish rigid governance over the network's process control plans. Every site must run identical tooling strategies, cutter paths, and cycle parameters for any dimension entering the RSS calculation. If a site needs to alter the manufacturing process to localise tooling or reduce cycle time, they must submit a new capability study. The central team recalculates the root-sum-square model using the new SPC data before approving the change. This discipline protects the statistical integrity of the submission.
Statistical tolerancing is only as strong as the weakest SPC chart in the manufacturing network.
The Cost Impact of Scale
The financial consequence of worst-case tolerancing multiplies exponentially across a manufacturing network. Forcing a ±0.02 mm tolerance on a dimension that naturally holds ±0.10 mm demands precision grinding at every facility. The piece price inflation hits the entire production volume. When a supplier runs three plants producing 100,000 units annually each, the decision to use worst-case math costs the supply chain millions. That capital funds unnecessary machine time, additional gauge complexity, and inflated scrap rates at facilities struggling to maintain the artificial precision.
Implementing RSS across the network recovers this margin. A rotor bearing assembly requiring seven distinct dimensions illustrates the scale of the savings. The worst-case stack-up demands ±0.021 mm on each component, forcing grinding across all sites. The RSS calculation yields an assembly variation of ±0.112 mm against a functional requirement of ±0.15 mm. Every facility can drop the grinding operation, return to standard machining, and meet the assembly requirement with statistical confidence. The cost differential between grinding seven components across three plants and standard machining across the same network represents immediate, recoverable operating margin.
This cost recovery requires the central quality team to sell the statistical argument internally before presenting it to the customer. Plant managers at the satellite facilities will resist widening tolerances if they have historically been punished for scrap. The quality leadership must demonstrate that the RSS model actually improves first-pass yield by aligning the drawing limits with the true process capability of the network. Once the plants see their scrap rates drop without impacting assembly function, the cultural resistance to statistical tolerancing dissolves.
Critical Thresholds for Multi-Site RSS Validation
Governance for the Engineering Change
Scaling statistical tolerancing across multiple sites requires a documented engineering standard that dictates exactly when to deploy RSS and when to retain worst-case analysis. The central quality team cannot leave this decision to individual plant engineers. The standard must specify that pure RSS is acceptable for general functional dimensions where the cost of failure is moderate. For safety-critical features governed by AS9100 or IATF 16949, the standard must mandate modified statistical methods or full worst-case analysis regardless of process capability.
The safety factor method provides a critical compromise for medium-risk assemblies spread across the network. Multiplying the RSS result by a coefficient of 1.5 narrows the predicted assembly tolerance, providing a deliberate buffer against process drift at less capable satellite facilities. This gives the customer a quantifiable margin of safety while still avoiding the extreme cost of full linear addition. Monte Carlo simulations offer another layer of governance for complex, non-normal distributions, allowing the central team to model thousands of assembly combinations based on aggregated SPC data.
The governance standard must also define the engineering change process for tolerance modifications. If a satellite site identifies an opportunity to widen a tolerance based on new SPC data, the standard dictates the path. The site submits the data to the central quality team, who recalculate the RSS model, verify the network impact, and approve the drawing change. This structured process ensures that statistical tolerancing decisions remain centralised, evidence-based, and defensible during any customer audit across the global supply chain.
Sustaining the Discipline Across the Network
A multi-site RSS strategy is not a one-time PPAP exercise. Equipment wears, tooling dulls, and process drift is inevitable across a network of manufacturing facilities. The central quality team must implement a sustaining audit programme that regularly pulls SPC data from every site to verify the underlying assumptions of the RSS calculation. If a satellite facility's Cpk drops below 1.33 on a critical dimension, the statistical model is compromised. The central team must halt the tolerance relaxation and address the process instability immediately.
This sustaining effort requires digital connectivity. Relying on plants to email static SPC charts to the central quality team monthly is a flawed, reactive system. The robust solution integrates the manufacturing execution systems of the various facilities into a centralised data lake. The quality team monitors real-time dimensional data, automatically flagging any special-cause variation that threatens the RSS calculation. This digital infrastructure provides the exact evidence the customer needs during a surveillance audit.
Statistical tolerancing fails at scale when the supply chain treats it as a mathematical trick to loosen drawings. It succeeds when the organisation treats it as a rigorous discipline that demands advanced SPC, strict process governance, and centralised quality control. The supplier that builds this infrastructure can confidently present RSS calculations to any customer, demonstrating deep process knowledge, protecting assembly function, and recovering the margin that worst-case engineering systematically destroys.