--- translationKey: humanoid-robot-parts-prototype-to-production lang: en slug: humanoid-robot-parts-prototype-to-production title: 'From Prototype to Production: CTQ, Tolerance Stack-Up and Process Control for Humanoid Robot Parts' description: 'How precision humanoid robot parts move from prototypes to stable production: define CTQs, build functional tolerance stack-ups, validate the production process and measurement system, then control capability, traceability and change.' publishDate: 2026-07-30 updateDate: 2026-07-30 draft: false featured: true category: quality-management industries: - humanoid-robot - general-manufacturing tags: - 'humanoid robot production launch' - 'CTQ management' - 'tolerance stack-up' - 'control plan' - 'SPC' - 'Cpk' - 'measurement system analysis' - 'process change control' author: 'Zhongde Precision Engineering Team' reviewedBy: 'Zhongde Precision Engineering Team' image: /images/knowledge-center/humanoid-robot-parts-prototype-to-production-en.webp imageAlt: Quality control chain from prototype to production for precision humanoid robot parts showArticleImage: false directAnswer: 'Moving from prototype to production is not a matter of copying a successful CNC program to more machines. Product function must first be translated into CTQs and functional tolerance chains, then into production equipment, fixtures, tooling, measurement methods, sampling frequencies, reaction plans and change control. One conforming prototype proves that the part can be made under one set of conditions; production release must prove that the process can repeat the result at the required rate through material-lot changes, tool wear, shift changes and downstream treatments.' relatedPages: - /en/humanoid-robot-joint-machining - /en/quality/ctq-management - /en/quality/tolerance-stack-up-management - /en/quality/manufacturing-process - /en/quality/inspection-equipment relatedArticles: - humanoid-robot-joint-actuator-housing-machining - humanoid-robot-joint-actuator-components-machining - humanoid-robot-lightweight-materials faq: - question: 'If every prototype dimension passes, why is the part not ready for production?' answer: 'Prototype work often receives extra setup time, measurement, manual correction and support from highly experienced personnel. It may not represent production rate, material-lot changes, tool life, shift changes, repeated clamping, subcontracted finishing or long-term drift. Production release therefore validates both the product and the process, not one inspected part.' - question: 'Who should define CTQs for humanoid robot parts?' answer: 'CTQs should be confirmed jointly by product design, assembly, manufacturing and quality. Design explains function and risk, assembly provides the real mating relationships, manufacturing identifies process-sensitive features, and quality defines the measurement and control method. A supplier should not infer CTQs only from the smallest tolerances.' - question: 'Can Cpk be calculated from a pilot lot of only a few dozen parts?' answer: 'A small pilot lot can provide preliminary statistical insight, but whether it is valid formal capability evidence depends on sample size, sampling method, process stability, distribution and customer requirements. A short sequence produced with repeated adjustments does not represent a long-term production process, so one Cpk value must not be used alone.' - question: 'After production starts, should only CTQs be inspected?' answer: 'No. CTQs require a higher level of control, but other dimensions, appearance, material, cleanliness and packaging still need risk-based verification. CTQ management prioritizes resources around function and failure risk; it does not cancel other requirements.' - question: 'Does changing the machine, fixture or surface-treatment supplier require revalidation?' answer: 'An impact assessment is normally required. Any change that may alter datums, distortion, dimensional compensation, roughness, coating thickness, measurement results or capability should trigger an appropriate level of first-piece inspection, pilot production, stack-up review, capability confirmation or customer approval.' --- ## Direct conclusion The change from prototype to production is not simply a change from 5 parts to 5,000 parts. The manufacturing objective changes: ```text Prototype: prove that the design can be made Pilot production: prove that the process route can be repeated Serial production: prove that the process remains controlled over rate, lots and time ``` A prototype engineer may use repeated alignment, manual offsets, additional measurements and conservative cutting conditions to complete one conforming part. A production process cannot depend on one person's memory or be re-tuned for every component. The knowledge must be converted into defined datums, programs, fixtures, tool-life rules, inspection methods, sampling frequencies and reaction plans. For humanoid robot joint housings, bearing carriers, reducer locating seats, output flanges and encoder mounts, a robust launch depends on three linked tasks: 1. **Identify the CTQs that truly determine joint function.** 2. **Organize individual tolerances into functional stack-ups.** 3. **Validate the process under production conditions and keep it controlled.**
Quality control chain from frozen requirements and CTQ stack-up through pilot validation, capability, production release, SPC and change control
Figure 1 | Production launch is not a simple increase in quantity. It is the conversion of functional requirements into a validated, traceable and continuously controlled manufacturing chain.
## 1. Why does a conforming prototype not prove production readiness? Prototype manufacturing commonly benefits from conditions that will not exist in regular production: - low quantities and generous setup time; - highly experienced operators; - selection of the best available machine and fresh tooling; - fixtures, programs and inspection methods still being adjusted; - individual correction after each deviation; - limited exposure to material lots, shift changes and long tool-life cycles. A prototype inspection report therefore proves only that the part met requirements under the material, machine, clamping, tooling, environment and measurement conditions used at that time. It does not automatically prove that: - the next material lot will distort in the same way; - another operator will reproduce the same datum setup; - a bore remains centered near the end of tool life; - anodizing variation will not change a bearing fit; - part temperature remains controlled at production rate; - coaxiality and face runout remain stable after hundreds of cycles. > **A prototype is initial evidence of product manufacturability. Production requires evidence for both the product and the manufacturing process.** ## 2. Freeze the requirements before creating control charts Before process data is analyzed, the team must define what is being repeated. Statistical tools cannot compensate for an unstable drawing revision, an undefined assembly condition or a disputed datum system. At minimum, confirm the following: | Item to freeze | Questions that must be answered | | -------------------- | ------------------------------------------------------------------------------ | | Drawing and revision | Which revision is valid? What governs when 2D and 3D data conflict? | | Functional datums | Which axis, face and pilot actually locate the assembly? | | Mating components | Are bearing, reducer, motor and encoder types and fits confirmed? | | Acceptance state | Is the feature accepted after machining, coating, press-fit or final assembly? | | Volume and rate | What are prototype, pilot, monthly and target cycle-time quantities? | | Required evidence | Are full dimensional, material, capability or assembly reports required? | | Change authority | Which changes require notification or customer approval? | Humanoid robot designs can evolve rapidly. Design change itself is manageable; uncontrolled propagation of the change into programs, fixtures, inspection plans and work-in-process inventory is the real risk. ## 3. CTQ is not a list of the tightest tolerances A CTQ, or Critical to Quality characteristic, has a decisive influence on function, assembly, performance, safety, life or customer failure risk. It may be dimensional, but it may also involve material state, surface condition, cleanliness, press force or functional test results. Typical CTQs for a joint actuator housing include: - the relationship between two bearing bores and the common joint axis; - reducer pilot coaxiality to the bearing axis; - output-face runout relative to the functional axis; - motor mounting-face flatness and perpendicularity; - encoder-mount eccentricity relative to rotation; - final bearing-bore size and cylindricity; - free-state distortion after unclamping; - precision interfaces after anodizing or other finishing; - assembled rotational torque, preload or clearance. Some small tolerances exist mainly for appearance or a traditional drafting practice. Other apparently generous tolerances contribute to a long chain and can strongly affect bearing preload, gear engagement or encoder air gap. CTQ selection should therefore combine: ```text Product function + assembly relationship + failure mode + contribution to the stack-up + process sensitivity + measurement feasibility ``` ## 4. Build the tolerance chain from the final function backward A tolerance stack-up shows how variation from multiple features and components accumulates at the final assembly function. ### 4.1 Axial stack-up example Joint output bearing preload or end clearance may depend on: ```text Housing shoulder location + bearing width + spacer thickness + reducer end-face position + cover clamping face + shim or locking component = final axial clearance or preload ``` Each part can be produced near its drawing nominal and the assembly can still become too tight or too loose if the complete chain was not checked. ### 4.2 Radial and coaxial stack-up example Output-flange radial runout may include contributions from: ```text Housing bearing-axis error + bearing clearance and press-fit distortion + output-shaft fit error + reducer output-interface error + flange pilot error = final output radial runout ``` ### 4.3 Three questions every stack-up should answer 1. What is the final functional requirement? 2. Which dimensional and geometric errors contribute to it? 3. Which contributors should be changed by design, and which should be controlled by manufacturing or assembly? The objective is not to tighten every tolerance. A better design allocates tolerance to stable, measurable and cost-effective features, while using shims, selective assembly or adjustment where appropriate. ## 5. Convert prototype findings into production documents The most valuable result of prototype work is not the conforming part itself, but the reusable process knowledge created while making it. For every deviation, determine: - Was the product definition unclear, or was the process unstable? - Did material stress, clamping, cutting, heat, finishing or measurement create the error? - Can the correction be standardized? - Will the problem return after a lot, tool, shift or machine change? The conclusions should flow into linked documents: | Document | Production purpose | | ------------------------ | --------------------------------------------------------------------------------------- | | Process flow diagram | Defines every step and sequence from incoming material to shipping | | PFMEA | Identifies failure modes, causes, effects and preventive controls by operation | | Control plan | Defines product/process characteristics, method, frequency, responsibility and reaction | | Work instruction | Converts alignment, clamping, tooling and setup knowledge into executable steps | | Inspection specification | Defines datums, part state, equipment, program and acceptance method | | Tool-life plan | Defines preventive replacement, compensation and abnormal-wear handling | | Traceability plan | Links material, equipment, program, fixture, operator and results | The process flow, PFMEA and control plan should use the same operation sequence and identifiers. Documents that describe different processes cannot control the shop floor. ## 6. Pilot production must represent production conditions The purpose of a pilot lot is not to produce another set of carefully managed prototypes. It is to expose the variation of the intended production system. Where practical, use: - the planned production location and machine; - production fixtures, soft jaws and gauges; - production tooling, programs and cutting parameters; - production material sources and conditions; - actual operators and shift arrangements; - planned inspection methods and sampling frequency; - real finishing, cleaning, assembly and packaging routes; - a cycle time close to the intended production rate. A pilot run performed with laboratory-speed machining, temporary fixtures, extra hand correction and exceptional 100% inspection cannot represent normal production. The German VDA production-process and product-approval approach emphasizes parts manufactured at the serial location with serial equipment, serial tooling, qualified personnel and the planned process flow. The same principle applies to robot precision parts: **the objective is not to demonstrate the peak accuracy of one machine, but the repeatable output of the full production system.** ## 7. Confirm the measurement system before evaluating capability CTQ data is useful only when the measurement system is sufficiently reliable. Common risks include: - different operators selecting different datums or points; - a CMM program that does not match the functional datum system; - a thin-wall part being distorted again by the inspection fixture; - different results for a warm part and a temperature-stabilized part; - roughness, roundness or bore form influencing contact readings; - disagreement between shop-floor gauges and laboratory equipment. Before formal SPC or capability analysis, confirm: 1. adequate resolution for the tolerance; 2. acceptable repeatability; 3. agreement between operators and equipment; 4. measurement datums aligned with product function; 5. defined environment and part condition; 6. calibrated and traceable equipment. MSA or Gage R&R is not paperwork for its own sake. It prevents measurement noise from being mistaken for manufacturing variation. ## 8. Cpk is not a production certificate Capability analysis evaluates whether the distribution of a stable process has sufficient margin within the specification. The main distinctions are: - **Cp** compares process spread with specification width; - **Cpk** also reflects how well the process is centered; - **Pp/Ppk** are often used for broader or longer-term performance, subject to the agreed method. A high Cpk can still be misleading when: - all data comes from one new tool and one shift; - operators repeatedly adjust the process during sampling; - samples are consecutive observations from a short period; - a clearly non-normal distribution is forced into a normal model; - the measurement system contributes excessive variation; - machine, fixture or material lot changes occur inside the dataset; - inconvenient data points are removed without technical justification. A more reliable sequence is: ```text Define the characteristic and measurement method → validate the measurement system → collect representative data → confirm process stability → select an appropriate statistical model → calculate and interpret capability → maintain monitoring and reaction plans ``` Bosch's public machine- and process-capability guidance similarly treats capability indices as estimates of the current condition. They require a stable, predictable process and an appropriate statistical model; one index cannot replace engineering judgement. ## 9. A control plan must say what happens after an abnormal result Many control plans list a dimension, gauge and frequency but do not provide a practical reaction. When a trend or violation occurs, the operator then has no clear instruction to stop, contain, offset or continue. An executable CTQ control item should include: | Control-plan field | Example | | ------------------------ | ---------------------------------------------------------------------------------------- | | Product characteristic | Output bearing-bore diameter and cylindricity | | Process characteristic | Fine-boring offset, spindle temperature, clamping pressure | | Specification and target | Drawing limit, process center and warning level | | Measurement method | Air gauge / CMM / roundness tester and program number | | Frequency | First part, after tool change, every defined quantity and last part | | Record and traceability | Machine, tool, fixture, lot and result | | Reaction plan | Stop, contain from last accepted point, remeasure, find cause and release before restart | The containment boundary must be explicit. When one nonconforming part is detected, the response cannot be limited to that single piece; all production since the last confirmed acceptable check requires evaluation. ## 10. Early production needs strengthened control The launch period is still learning the actual variation of the process. It normally requires stricter controls than mature production: - increased CTQ sampling; - first-part, tool-change, shift-change and last-part confirmation; - additional checks after unclamping, press fitting or surface treatment; - tracking of tool life and compensation changes; - comparison of new material lots; - more assembly or functional verification; - daily review of defects, rework and trends; - reduction of inspection only after defined exit criteria are met. Exit criteria should not be merely “no defects for several days.” They should cover process stability, measurement reliability, agreed capability, effective reactions and controlled changes. ## 11. Which process inputs matter for humanoid robot parts? SPC does not have to monitor only final dimensions. For CTQs that are delayed, expensive or difficult to measure, process inputs that predict the result may also need control. ### Joint housings and bearing carriers - material condition and lot; - distortion after roughing; - support position and clamping pressure; - fine-boring tool life and offsets; - machine and part temperature; - free-state bore size, form and coaxial relationship; - anodizing thickness and masking condition. ### Output flanges and shafts - pilot and face runout; - heat-treatment lot and hardness; - datum transfer between turning and grinding; - bearing-seat roundness and roughness; - spline or tooth relationship to the main axis. ### Encoder and sensor mounts - mounting-face flatness; - dimensions contributing to sensing gap; - eccentricity to the rotational center; - locating-hole position; - cleanliness, burrs and assembly damage. A process input should be controlled because it has a demonstrated engineering relationship to the output, not merely because it is easy to collect. ## 12. Traceability is more than a batch label Effective traceability should answer: > Which material lot, machine, program, fixture, operator, time, tooling set, finishing lot and measurement record produced this batch? A useful chain is: ```text Order and part revision → material heat/lot → blank and machining lot → machine, program, fixture and tool → inspection record → surface-treatment or subcontract lot → assembly, packaging and shipment lot ``` The purpose is not only to assign responsibility after a failure. Traceability allows a team to reduce the affected scope quickly instead of isolating all inventory and work in process. ## 13. Freeze the approved process without blocking improvement ABB's Swiss supplier-quality framework includes formal control of selected process changes after production approval. The principle is not to prevent improvement, but to stop unvalidated change from damaging demonstrated capability. Typical changes for humanoid robot precision parts include: - material grade, temper, source or blank form; - machine, production site or subcontractor transfer; - program, operation sequence or clamping change; - cutting-tool brand, geometry or life rule; - anodizing, heat-treatment or cleaning condition; - CMM program, gauge or inspection method; - fixture repair that changes location; - increased rate or different lot size. The validation level should reflect impact on CTQs, the stack-up and capability: ```text Document update < first-piece confirmation < pilot lot < capability review < assembly / functional validation < customer reapproval ``` ## 14. Recommended prototype-to-production workflow ### Stage 1: Prototype manufacturability - Review drawing, datums, material and functional interfaces. - Produce and fully inspect the first parts. - Record distortion, tooling, clamping and finishing issues. - Confirm design deviations and assembly feedback with the customer. ### Stage 2: CTQ and tolerance-chain confirmation - Identify CTQs from joint function and failure modes. - Build axial, radial and assembly stack-ups. - Define final acceptance state and measurement method. - Decide which characteristics require capability or 100% functional verification. ### Stage 3: Production process planning - Select production equipment, fixtures, tools and programs. - Complete process flow, PFMEA and control plan. - Define tool life, sampling and abnormal reaction. - Establish material and process traceability. ### Stage 4: Pilot production under production conditions - Use actual operators, equipment, fixtures and rate. - Validate capacity, tool change, shift change and logistics. - Collect CTQ and key-process-parameter data. - Verify final condition after finishing and assembly. ### Stage 5: Measurement and capability validation - Confirm critical measurement systems. - Evaluate process stability. - Assess capability using the customer-agreed method. - Center the process, reduce variation or add error-proofing where needed. ### Stage 6: Production approval and safe launch - Complete first-piece, dimensional, material and process evidence. - Assign temporary controls to any open risk. - Start with increased sampling and daily review. - Move to normal control only after the exit criteria are met. ### Stage 7: Ongoing control and change management - Monitor drift using SPC, first/last-part checks and audits. - Control changes to tools, fixtures and measurement programs. - Contain abnormal production and implement permanent corrective action. - Feed production data back into the stack-up, process and design. ## 15. What information should be provided for RFQ and production launch? For a realistic production-risk assessment, provide: 1. current 2D drawing, 3D model and revision history; 2. component location and function within the joint assembly; 3. CTQ, special-characteristic or critical-interface identification; 4. bearing, reducer, motor and encoder information; 5. stack-up, assembly clearance, preload and runout requirements; 6. material grade, condition, blank form and surface treatment; 7. prototype findings, assembly feedback and design-change history; 8. prototype, pilot, monthly and lifetime quantities; 9. required capability, dimensional, material and functional reports; 10. packaging, cleanliness, traceability and change-approval requirements. When product function, manufacturing process and production requirements are placed on the same engineering map, the supplier can distinguish which risks must be solved by machining accuracy and which require design, assembly, selective fitting, inspection or process control. ## Conclusion: production capability means stable repetition Production of humanoid robot precision parts is not achieved by copying a prototype program to more machines, nor by increasing inspection on every dimension. A mature launch creates a closed loop: ```text Functional requirement → CTQ → tolerance chain → process and risk planning → pilot under production conditions → measurement and capability validation → control plan and SPC → traceability, reaction and change control → data fed back to design and process ``` In projects involving joint housings, bearing carriers, output flanges, reducer seats and encoder mounts, Zhongde looks beyond whether prototype parts can be shipped. We also evaluate whether production equipment and fixtures are defined, CTQs are measurable, tolerance chains close correctly, tooling and finishing variation are controlled, and any future change can be assessed quickly. Production must prove not merely that the part was made correctly once, but that the team understands why it was correct and can continue to reproduce it. ## Engineering references - [VDA 2: Production Process and Product Approval](https://vda-qmc.de/en/education/410-vda-2-produktionsprozess-und-produktfreigabe-ppf/) - [AIAG Quality Core Tools: APQP, Control Plan, PPAP, FMEA, MSA and SPC](https://www.aiag.org/expertise-areas/quality/quality-core-tools) - [Bosch: Machine and Process Capability](https://assets.bosch.com/media/global/bosch_group/purchasing_and_logistics/information_for_business_partners/downloads/quality_docs/general_regulations/bosch_publications/booklet-no09-maschinen-und-prozessfaehigkeit_de.pdf) - [ABB: Global Supplier Quality Manual](https://new.abb.com/motors-generators/supplying) - [Hexagon Q-DAS: Process Capability and SPC](https://hexagon.com/products/product-groups/manufacturing-project-management/statistical-process-control-software) ## Frequently asked questions ### If every prototype dimension passes, why is the part not ready for production? Prototype work often receives extra setup time, measurement, manual correction and support from highly experienced personnel. It may not represent production rate, material-lot changes, tool life, shift changes, repeated clamping, subcontracted finishing or long-term drift. Production release therefore validates both the product and the process, not one inspected part. ### Who should define CTQs for humanoid robot parts? CTQs should be confirmed jointly by product design, assembly, manufacturing and quality. Design explains function and risk, assembly provides the real mating relationships, manufacturing identifies process-sensitive features, and quality defines the measurement and control method. A supplier should not infer CTQs only from the smallest tolerances. ### Can Cpk be calculated from a pilot lot of only a few dozen parts? A small pilot lot can provide preliminary statistical insight, but whether it is valid formal capability evidence depends on sample size, sampling method, process stability, distribution and customer requirements. A short sequence produced with repeated adjustments does not represent a long-term production process, so one Cpk value must not be used alone. ### After production starts, should only CTQs be inspected? No. CTQs require a higher level of control, but other dimensions, appearance, material, cleanliness and packaging still need risk-based verification. CTQ management prioritizes resources around function and failure risk; it does not cancel other requirements. ### Does changing the machine, fixture or surface-treatment supplier require revalidation? An impact assessment is normally required. Any change that may alter datums, distortion, dimensional compensation, roughness, coating thickness, measurement results or capability should trigger an appropriate level of first-piece inspection, pilot production, stack-up review, capability confirmation or customer approval.