Key takeaways:
Binding reference standard: In production, an approved good part defines the target state as a benchmark. In quality control, a reference set consisting of good, borderline, and defect samples often complements this standard in order to ensure a robust camera setup and inspection design.
Clear alignment with suppliers: A shared reference part prevents misunderstandings, as expectations and borderline cases are evaluated based on the exact same standard.
Clean approval process: The Golden Sample is only created after testing, adjustment, and formal approval, making it distinct from a prototype or an initial sample.
Foundation for AI inspections: Golden samples represent the digital inspection benchmark used to validate hardware and cameras, ensuring that defects are visible in the image and can be recognized by the AI.
Stable application in series production: Clear decision-making processes, regular calibration, and a well-maintained catalog of borderline cases keep evaluations consistent and prevent unnecessary rejections.
A Golden Sample is an approved reference standard that defines the target state of a component. It serves as a benchmark for evaluating all subsequently produced parts, ensuring that decisions in quality control remain consistent and reproducible. In practice, this benchmark is frequently supplemented by examples covering typical borderline cases—i.e., feature values close to the acceptance limit. Depending on the field, the term Golden Sample is understood in different ways:
in production: An approved good part represents the target state and serves as a reference standard for series production.
in quality assurance: A reference set combines the good part with selected defective parts for each defect class, including borderline cases. This allows the camera setup and lighting to be designed so that relevant defect types and borderline cases are reliably visible and assessable.
As a shared point of reference between the manufacturer, supplier, and quality team, the Golden Sample is linked to specific inspection characteristics and inspection instructions, for instance, for visual inspection in quality control. Depending on the component, additional non-destructive testing (NDT) methods may be useful if relevant characteristics cannot be reliably assessed from the outside.
Without a shared benchmark, quality inspection quickly leaves room for interpretation: What is still acceptable, and what is scrap? A reference part set bridges this gap. It makes evaluations comparable across lines, locations, and suppliers.
Where the reference part set provides concrete support:
Supplier communication: Instead of abstract requirements, the Golden Sample provides a clear reference standard for how the part must look and function. This shortens coordination processes and prevents two parties from having different perceptions of quality.
Approvals & ongoing inspection: At the start of production, during variant changes, or in borderline cases, the reference part set serves as a decision-making aid. Inspection criteria are aligned with it, reducing recurring discussions.
Audits & complaints: If a complaint arises, the evaluation can be explained transparently based on the reference benchmark. Which characteristics deviate, and why is this relevant?
Process optimization: Deviations are identified, clearly described, and consistently recognized. This makes root cause analyses and measures applied to tools, parameters, and material batches more targeted.
This benchmark is particularly crucial for automated inspection lines: An AI quality control system operates more stably when it is clear what "good" is measured against. For deviations that are not registered as classic errors, data-driven anomaly detection complements the reference inspection effectively.
A Golden Sample is created through an approval process that closely integrates manufacturing and quality. This typically begins with a production-ready part that is inspected based on defined characteristics. If deviations occur, the process or tool is adjusted. If inspection criteria are unclear, they are defined more precisely. This cycle is repeated until the result consistently matches the expected quality and can be formally approved.
Clean reference part management after approval
The approval process typically involves engineering, quality assurance, and, in the case of purchased parts, the supplier. To ensure this process does not fail due to media disruptions, early digital management is key. Via PLM/PDM and BOM data, references, revisions, and associated inspection criteria can be cleanly linked – a central component of digitalization in mechanical engineering. In addition, clear rules for documentation and storage are needed so that the reference part can truly serve as a benchmark in day-to-day operations:
Labeling & versioning: Variant, revision, and date are uniquely defined, and the reference part is linked to the current drawing version.
Approval status & inspection characteristics: Who approved what and when is documented. Relevant inspection characteristics and instructions are also stored in a traceable manner.
Storage & handling: The Golden Sample is protected from damage and contamination. Clear rules exist for access, transport, and use.
For subsequent use in inspection processes, consistent visual documentation is also helpful: Methods of industrial image processing allow approved characteristics to be clearly mapped, for example, via reference images, cropouts, or defined inspection areas. This transforms the physical sample into a uniquely referenceable standard that can be consistently compared in automated inspection workflows.
A Golden Sample makes it possible to transfer quality knowledge from manufacturing into a digital inspection system. To ensure that AI reliably detects deviations, the reference state is defined concretely and stored as a data basis: Which characteristics are inspected, where exactly on the component or in the image are they evaluated, and which variations are still acceptable? The key here is that the target state is not just described, but exists as a verifiable reference state within the image.
The range of good parts, borderline cases, and defect samples is crucial. In practice, a curated set of representative examples for each defect class and corresponding good parts is typically used for this purpose. This set helps to adjust the camera setup, including lighting, so that defects are reliably visible in the image. At the same time, inspection areas and borderline rules are derived from these examples, ensuring that acceptance limits do not remain implicit but can be applied reproducibly. This creates a reproducible inspection standard that can be applied automatically, for example, in end-of-line inspections.
During operations, the benefit becomes apparent in more stable decisions, as parts are evaluated according to the same criteria regardless of who inspects them or at which station. At the same time, real-time monitoring allows accumulations of deviations to be detected early, for example, if certain errors increase significantly within a few minutes. This helps to narrow down root causes quickly and reduce scrap before a defect pattern spreads through production.
In series production, a Golden Sample often fails due to a lack of operational discipline: Who makes the final decision on borderline cases? How are evaluations kept in sync across lines and locations? And how do you prevent a different quality standard from creeping in over the weeks? These very points are the most common drivers of Pseudo scrap or an increasing false acceptance rate.
Rules for stable decisions:
Decision-making authority: It is defined who makes the final decision in borderline cases (e.g., Quality vs. Production) and how this decision is documented.
Calibration: There are regular alignment rounds in which multiple inspection stations evaluate the same examples and actively resolve deviations.
Borderline case catalog: Borderline cases are maintained as concrete reference cases (Good/Bad with justification) so that they do not have to be renegotiated every time.
Escalation path: A clear process exists for uncertainties (e.g., quarantine → second evaluation → root cause check → decision) instead of ad-hoc discussions on the line.
Effectiveness: A few but meaningful KPIs are tracked to detect drift early. Depending on the process, this could be the proportion of borderline cases, the rate of manual re-evaluations, or complaints in relation to internal quarantines.
This turns the Golden Sample into a stable decision-making tool in series production. The added value arises primarily because borderline cases no longer depend on the individual, location, or daily form, but are managed as a repeatable decision logic. This is precisely what reduces friction, avoids unnecessary rejections, and at the same time protects against overly lenient approvals.
A Golden Sample defines the approved reference standard and makes quality decisions in series production reliable – from inspection and approvals to collaboration with suppliers. As a shared point of reference, it helps with borderline cases, complaints, and process drift, as deviations can be consistently evaluated and clearly justified.
In combination with AI-supported quality control, the reference part becomes the foundation for consistent and reliable inspection. In quality control, this often refers to more than just a single good reference part: it frequently includes a reference set of borderline cases and defect samples that secures the camera setup and inspection logic. Keeping the reference status, borderline cases, and changes consistently maintained remains crucial.
