World Sleep Industry ENCYCLOPEDIA

Quality Control Testing

6.1.3

Definition, purpose, and scope

Quality-control testing is the controlled sampling, measurement, observation, and evaluation of raw materials, components, work in process, finished goods, packaging, or production environments against requirements established in advance. It supplies evidence for receipt, continued processing, rework, release, quarantine, or investigation and can reveal changes in suppliers, equipment, personnel, and processes over time. Sleep-related products include textiles, mattresses and furniture, electronics and wearables, fragrances and personal care products, foods, software, and combined products. Their safety characteristics, regulatory routes, destructive tests, and data forms differ, so one universal inspection checklist is not suitable for all of them.

Testing is one link in a quality system, not the sole source of quality. When design requirements are ambiguous, a process is unstable, or a supply chain is inconsistent, more end-of-line inspection usually detects only some defects; it cannot inspect an incapable process into conformity. Quality assurance establishes the system and preventive mechanisms, while quality control obtains and reviews evidence of conformity. They are related but not interchangeable. Internal inspection, supplier testing, third-party testing, type testing, regulatory certification, and clinical or efficacy research have different purposes. Passing one activity does not automatically establish conformity at another stage, in another lot, throughout shelf life, or for an actual sleep outcome.

Requirement baseline, risk classification, and control plan

Before testing begins, the organization should freeze requirements that can support a decision. These include product and material identity, drawing or formula revision, critical quality and safety characteristics, units, tolerances or limits, visual-defect classes, conditioning, method, sampling unit, acceptance and rejection rules, and authority to approve results. Each requirement should trace to applicable law, regulation, standard, customer specification, design output, or risk control, with its edition, effective date, and market scope identified. Marketing terms such as “premium,” “comfortable,” or “sleep promoting” do not become acceptance criteria unless they have a measurable definition, comparator, and evidence boundary.

The control plan allocates resources according to the consequence of error and expected process variation. Characteristics related to electric shock, fire, entrapment, suffocation, chemical exposure, microorganisms, dose, or software safety generally need stronger prevention, verification, and release controls than ordinary appearance. Calling a characteristic critical does not mechanically require one hundred percent testing of every item. Severity, occurrence, detectability, historical capability, destructive testing, and regulatory obligations jointly inform supplier controls, incoming inspection, first-article checks, patrol inspection, online monitoring, final inspection, and periodic confirmation. A design or process change should trigger reassessment so an obsolete specification is not silently applied to a changed product.

Lot definition, sampling, and sample chain

Sampling starts with a lot boundary and sampling frame. The plan should state whether a lot shares material, formula, machine, mold, shift, date, or software configuration. Samples should be selected randomly or by justified strata so edge locations, startup periods, packaging layers, tool cavities, roll segments, or pallets are not systematically missed. Record the sampler, time, location, quantity, and seal. Convenience samples, laboratory submissions, and supplier-selected showpieces may not represent the lot. Composite samples can reduce cost but may hide a localized exceedance. Destructive, scarce, or expensive tests require a risk-based statistical balance between sample size and the chance of missing a defect.

Acceptance sampling primarily decides whether to accept or reject the current lot. It does not estimate the lot's exact defect rate and does not prove zero defects. The scheme should state whether decisions use attributes or variables, the sample size, acceptance and rejection numbers, switching rules for tightened or reduced inspection, producer and consumer risks, and treatment of inconclusive results. Even one hundred percent visual inspection is affected by fatigue, visibility, and classification consistency and therefore does not inherently guarantee perfect detection. From withdrawal through transport, conditioning, opening, subdivision, preparation, testing, retention, and disposal, every sample needs a unique identity, status, and custody history to prevent contamination, substitution, moisture loss, volatilization, software overwrite, or return of a tested item to normal inventory.

Methods, personnel, and measurement-system capability

A method should fit the characteristic, matrix, range, resolution, and intended decision. When a standard method is used, its edition, scope, and permitted options need review, and the laboratory should verify correct performance with its own equipment, personnel, and samples. A modified method, a new matrix, or a laboratory-developed procedure requires validation proportionate to its effect. Relevant characteristics can include selectivity, trueness or recovery, repeatability, reproducibility, linearity, working range, detection and quantitation capability, robustness, and sample stability. Personnel require training and authorization. Subjective appearance or sensory decisions benefit from defect atlases, boundary samples, and checks of agreement among inspectors.

A calibration label on an instrument does not by itself make a result reliable. The measurement system also includes reference materials, fixtures, software, environment, operator, preparation, and calculations. Metrological traceability is a property of a measurement result and depends on a documented unbroken calibration chain with associated uncertainties. Equipment should cover the intended range and be subject to intermediate checks, maintenance, and out-of-tolerance action. Gauge repeatability and reproducibility, bias, linearity, stability, and resolution should be appropriate to the tolerance and risk. Laboratory competence must match the specific test and method scope. General accreditation or one proficiency test does not replace continuing internal controls, blind samples, duplicates, reference materials, and interlaboratory comparisons.

Incoming, in-process, final, and digital inspection

Incoming control should connect material name and specification, supplier, manufacturing lot, certificate, and physical status. A supplier certificate can justify reduced duplicate testing only after supplier qualification, confirmation of sampling and methods, verification of report authenticity, and continuing performance review. High-risk material may still require identity testing, periodic full-panel confirmation, or independent analysis. In-process inspection is intended to expose drift promptly and should link results to equipment, tooling, formulation, operator, shift, and environment. Control limits describe statistical process behavior, while specification limits describe product requirements. A statistically stable process can still be incapable of meeting specifications, and one result inside specification does not show that the process is stable.

Final inspection should represent the delivered assembly, firmware, accessories, packaging, and conditioning and should verify labels, instructions, traceability codes, and critical safety functions. Automated vision, sensors, test scripts, and artificial-intelligence models can expand coverage, but their training samples, thresholds, versions, false-positive and false-negative behavior, permissions, and human-review rules also require control. A software upgrade or change in camera position or illumination can invalidate prior verification. Electronic raw data, images, spectra, and logs should preserve time, sample, equipment, program version, and audit trail. Keeping only an aggregate “pass” while discarding failures, reruns, manual changes, or missing data is not an adequate record.

Atypical results, nonconformity, retesting, and lot disposition

When an atypical result, adverse trend, or out-of-specification result appears, the original sample, raw data, and affected lot should be secured immediately. Identity, sampling, environment, reference materials, equipment, calculations, method execution, and data transfer are reviewed before deciding whether evidence supports an assignable laboratory cause. A result may be invalidated only when the cause is identified and documented under procedure. A result that is merely unexpected, a passing result from a different portion, or one passing repeat is not a reason to delete the original. Triggers, number of repetitions, statistical combination, and conclusion rules for retesting, resampling, or expanded sampling should be defined in advance. Testing until a pass appears and arbitrary averaging both conceal rather than resolve variation.

Authorized personnel make disposition decisions from the complete evidence. Options can include acceptance, concession, sorting, rework, downgrade, return, scrap, production hold, or an expanded investigation. A concession cannot bypass a statutory, safety, or critical-performance requirement and should identify its lot, quantity, use, approver, and validity period. The investigation evaluates the same lot, adjacent lots, shared materials and equipment, and product already shipped, and initiates customer notification, corrective action, or recall where necessary. After sorting or rework, affected characteristics require reconfirmation; a repair cannot be assumed free of new contamination, damage, software configuration errors, or service-life risk.

Reporting, release, change, and continual improvement

A complete record commonly includes request and purpose, sample and lot identity, sampling plan and custody, specification and edition, method and deviations, staff authorization, equipment and calibration status, environment, raw data, calculations, quality controls, measurement uncertainty, decision rule, atypical-result investigation, disposition, and review signatures. The report should separate observations, conformity decisions, and scope and should identify subcontracted work, untested attributes, detection or quantitation limits, and deviations. Near a limit, rounding, uncertainty, and the risks of false acceptance and false rejection should follow the agreed decision rule rather than preserving only a favorable word such as “pass.”

Release confirms that required tests, deviations, investigations, and records are complete and is approved by personnel with defined authority or appropriate independence from test execution. A change in material, supplier, site, equipment, tooling, formulation, software, method, limit, packaging, or intended use requires assessment of whether earlier evidence remains applicable and what verification must be repeated. Defects, scrap, complaints, returns, repairs, and field events should be trended by lot and cause and fed back to design, purchasing, and process improvement. Quality-control testing establishes only the result for defined samples, conditions, and requirements. By itself, it cannot guarantee that every product will never fail, grant market authorization, establish suitability for every user, or demonstrate an improvement in sleep.