Tolerance Stack-Up Analysis in Mechanical Assemblies

Every assembly is a stack of tolerances. The clearance that the drawing promises is really the sum of the machining variations, and the sum is what decides whether the parts fit. Tolerance stack-up analysis is the calculation that predicts the fit before the prototype, and it is the tool that prevents the assembly-floor surprises. This article covers the methods, the inputs, and the habits that keep the stack analysis honest.

Why the Stack Matters

The simple assembly looks forgiving. The shaft slides into the bearing, the bearing sits in the housing, and the housing bolts to the frame. The drawing shows the nominal dimensions and the clearance, and the assembly looks easy. The reality is that every dimension in the chain carries a tolerance, and the tolerances add.

The worst-case stack is the sum of all the tolerances in the chain. The worst case happens when every dimension is at its extreme in the direction that closes the gap. The worst case is rare in practice, but the design that fails the worst case is the design that fails sometimes.

The statistical stack is the more realistic view. The tolerances are treated as the distributions, and the sum follows the normal distribution. The statistical stack gives the probability of the fit, and the design that passes the statistical check with the margin is the design that fits in production.

The stack analysis answers the questions that the drawing alone cannot: the gap that remains at the worst case, the interference that occurs at the worst case, and the probability of the problem.

The Chain and the Datum

The first step is the definition of the chain. The chain is the sequence of the dimensions from the reference point to the feature of interest. The chain runs through the parts, the gaps, and the interfaces.

The chain should use the datums that the manufacturing uses. The dimension that is machined from the datum face is the dimension in the chain. The dimension that is measured in the inspection is the dimension in the chain. The chain that matches the process is the chain that predicts the real part.

The common mistake is the chain that skips a link. The designer follows the nominal dimensions on the drawing and misses the interface gap or the floating feature. The skipped link makes the analysis optimistic, and the optimistic analysis is the analysis that fails on the floor.

The Worst-Case Method

The worst-case method is the simplest and the most conservative. Each dimension in the chain contributes its full tolerance, and the total is the sum of the absolute values.

The worst-case gap is the nominal gap minus the total tolerance. The worst-case interference is the nominal interference plus the total tolerance. The design that passes the worst case is the design that always fits.

The worst-case method is used where the failure is unacceptable: the press fit, the safety critical clearance, and the interchangeability requirement. The worst-case method is also used when the tolerance chain is short, because the short chain makes the worst case close to the statistical case.

The cost of the worst-case method is the tight tolerance. The worst case with the long chain demands the tight tolerances, and the tight tolerances cost money. The designer that uses the worst case for every chain is the designer that overprices the part.

The Statistical Method

The statistical method, also called the RSS method, treats the tolerances as the independent random variables. The total variation is the square root of the sum of the squares of the tolerances.

The statistical method gives the smaller total for the same tolerances. The six-sigma total is the value that covers 99.7 percent of the assemblies, and the remaining 0.3 percent is the risk.

The statistical method is valid when the processes are in control and the dimensions are independent. The processes that drift, the fixtures that wear, and the dimensions that correlate all break the assumption. The statistical method that is used without the process knowledge is the method that is wrong by the hidden factor.

The hybrid approach is common in practice. The critical dimensions are treated with the worst case, and the non-critical dimensions are treated with the statistics. The hybrid gives the realistic total without the full risk of the pure statistics.

The Inputs That Decide the Result

The stack analysis is only as good as the inputs, and the inputs come from the process, not from the wish. The tolerance that is typed from the standard table is the input that has no relationship to the actual machine.

The process capability is the honest input. The milling operation holds plus or minus 0.05 on the normal day, and the grinding operation holds plus or minus 0.01. The tolerance that matches the process capability is the tolerance that the shop can hold without the heroics.

The inspection method matters too. The dimension that is measured on the CMM and the dimension that is measured with the caliper have the different uncertainties. The uncertainty of the measurement should be small compared to the tolerance, or the inspection decides the fit instead of the machining.

The temperature and the environment enter the analysis for the precision assemblies. The aluminum part that grows with the temperature and the steel part that stays put change the clearance by the difference of the expansion coefficients. The precision machine that is assembled in the climate-controlled room and used in the factory floor has the temperature stack that the analysis must include.

The Software and the Spreadsheet

The stack analysis is done in the spreadsheet or the dedicated software. The spreadsheet is the fast start for the simple chains. The dedicated software handles the complex chains with the Monte Carlo simulation and the sensitivity analysis.

The Monte Carlo simulation is the powerful statistical method. The dimensions are sampled from the assumed distributions, the chain is computed thousands of times, and the result is the distribution of the gap. The Monte Carlo handles the non-normal distributions and the correlated dimensions that the RSS method cannot.

The sensitivity analysis shows which dimensions matter. The dimension that contributes the most to the total variation is the dimension that should be tightened or redesigned. The sensitivity ranking is the guide for the tolerance allocation.

The Tolerance Allocation

The stack analysis is not only the verification; it is the allocation tool. The design that needs the total gap of 0.1 with the chain of four dimensions allocates the tolerance among the dimensions.

The allocation should follow the process capability and the cost. The dimension that is cheap to hold tight gets the tight tolerance. The dimension that is expensive to hold tight gets the loose tolerance. The allocation that balances the cost is the allocation that the shop can meet.

The common practice is the proportional allocation. The tolerances are allocated in the proportion of the process capabilities. The result is refined with the sensitivity analysis and the cost data.

The Habits That Keep It Honest

The stack analysis is a habit, and the habit has the rules. The first rule: analyze every critical assembly, not only the ones that are asked for. The second rule: use the process-based inputs, not the table values. The third rule: check the analysis against the prototype, and correct the inputs from the measurement.

The analysis that is never checked against the real parts is the analysis that builds the false confidence. The first article measurement is the validation of the stack. The measured gaps that match the predicted distribution confirm the analysis. The measured gaps that do not match send the team back to the inputs.

The documentation is the final habit. The stack analysis sheet with the chain, the inputs, the assumptions, and the result is the record that the next engineer can read. The undocumented analysis is the analysis that is redone from scratch.

Conclusion

Tolerance stack-up analysis turns the drawing promises into the assembly reality. Define the chain from the process datums, run the worst case for the critical fits and the statistics for the production view, use the process-based inputs, allocate the tolerances by the cost and the capability, and validate against the first articles. The assembly that is stacked honestly is the assembly that fits the first time, and the first-time fit is the schedule that stays on track.

A Worked Example: The Bearing Housing Stack

A bearing housing assembly shows the method in practice. The assembly has a base plate, a housing, a bearing, and a shaft. The critical requirement is the shaft clearance in the housing bore: the shaft must rotate freely, and the clearance must not exceed the allowed value.

The chain runs from the base datum through the housing height, the bore position, and the shaft diameter. The nominal dimensions and the tolerances are listed: the housing height with the milling tolerance, the bore position with the drilling tolerance, and the shaft diameter with the turning tolerance.

The worst-case calculation sums the tolerances. The clearance at the nominal is 0.1 millimeters, and the total tolerance is 0.3 millimeters. The worst-case clearance is minus 0.2 millimeters, which is an interference. The worst case says that the assembly can bind.

The statistical calculation gives the different view. The RSS total is the square root of the sum of the squares, and the value is 0.17 millimeters. The clearance at the six-sigma is the nominal minus the RSS total, which is minus 0.07. The statistical view says that the binding is rare but possible.

The design decision follows. The critical dimension, the housing height, is tightened to the grinding tolerance. The new total is 0.2 millimeters in the worst case, and the clearance stays positive. The cost of the grinding on the one dimension is accepted, and the assembly fits.

The example shows the three lessons: the worst case finds the risk, the statistics find the probability, and the allocation fixes the problem at the dimension that matters.

The Stack in the Design Process

The stack analysis belongs at two points in the design process. The first point is the concept stage, where the rough tolerances are assigned and the feasibility is checked. The second point is the detail stage, where the final tolerances are confirmed.

The concept-stage stack uses the standard tolerances and the generous margins. The stack that fails at the concept stage sends the design back to the concept: the different arrangement, the different datum, or the different adjustment.

The detail-stage stack uses the final tolerances and the process capabilities. The stack that passes at the detail stage releases the drawing with the confidence. The stack that fails at the detail stage triggers the tolerance allocation, the datum change, or the added adjustment.

The stack is also part of the change review. The tolerance change that is requested by the shop is checked against the stack before the approval. The change that passes the stack is approved quickly, and the change that fails the stack is rejected with the reason.

The Adjustments and the Shims

The design that cannot meet the stack with the machining alone uses the adjustment. The shim, the adjustable cam, and the threaded adjuster are the devices that absorb the tolerance stack.

The shim is the simplest adjustment. The gap is measured, and the shim of the required thickness is inserted. The shim is the standard solution for the preload and the clearance in the assemblies that are adjusted once.

The adjustable feature is the better solution for the repeated adjustment. The eccentric cam that adjusts the belt tension, the threaded stud that adjusts the height, and the wedge that adjusts the position are the features that are adjusted without the disassembly.

The adjustment adds the cost and the maintenance. The adjusted assembly needs the procedure and the documentation. The design that uses the adjustment where the fixed stack would fail is the design that trades the machining cost for the assembly time.

The Stack Tools in the CAD

The modern CAD systems include the tolerance analysis tools that read the model and compute the stack. The tools handle the 1D and the 3D chains, the worst case and the statistics, and the sensitivity.

The CAD-based stack is fast and consistent. The model is the source of the chain, and the tool finds the contributors. The result is the distribution, the sensitivity ranking, and the contribution percentages.

The CAD-based stack is still an input-driven analysis. The tolerances and the process assumptions are the inputs, and the garbage-in rule applies. The tool that is fed with the wish tolerances produces the wish result.