Food Science & Technology · Packaging, Safety, Quality & Regulation · also Food Quality Control, Quality Assurance, Total Quality Management in Food Industry, Laboratory Quality Management
Controlling quality in the plant and in the laboratory: quality systems and TQM, control charts and process capability, acceptance sampling and two- and three-class sampling plans, laboratory quality management and ISO/IEC 17025 accreditation, method validation and proficiency testing.
In the exams: ICAR AICE 35.a Unit 4 (quality systems, control charts, acceptance sampling) and 35.b Units 1 and 3 (ISO 17025, NABL accreditation, 2- and 3-class sampling plans); food analyst examinations
Quality in a food business means making, every time, a product that is safe, legal, true to its specification and what the customer expects. Quality assurance is the set of planned activities that gives confidence it will be so, and much of it rests on numbers: whether a filling line is drifting, whether a process can meet its specification, whether a lot should be accepted on the evidence of a sample, and whether a laboratory result can be trusted enough to reject a consignment or prosecute a supplier. A food technologist in quality or in a laboratory must understand both the management systems that organise this work and the statistics and metrology that make it objective.
This subject covers concepts of quality and quality systems, the tools of total quality management, and the ISO 9001 quality management system; statistical process control with control charts for variables and attributes and process capability indices; acceptance sampling, operating characteristic curves and the two- and three-class sampling plans used for microbiological criteria; quality control of raw materials, processes and finished products, including non-destructive and online inspection; and the laboratory side: laboratory quality management and ISO/IEC 17025, method validation, measurement uncertainty, proficiency testing and reference materials, and the accreditation bodies that recognise competent laboratories. The food safety management systems that sit alongside these (HACCP, ISO 22000 and certification schemes) are in food-safety-management-systems, and the underlying statistics in statistics-experimental-design-food.
These topics are examined in postgraduate and research-fellowship examinations in food safety and quality, which ask about quality systems, control charts, acceptance sampling, two- and three-class plans and laboratory accreditation; in food analyst and laboratory qualification examinations; in professional auditor and quality-engineer certifications; and in food technologist licensure examinations. The practice questions follow those patterns and give full working for control limits, capability indices, acceptance probabilities, z-scores and uncertainty budgets.
Quality has been defined in several ways, each useful. Juran called it fitness for use; Crosby, conformance to requirements; the ISO 9000 vocabulary defines it as the degree to which a set of inherent characteristics of an object fulfils requirements. For food, those requirements cover safety (freedom from hazards), legality (composition, labelling, additives and contaminants within the law of the market), nutrition, sensory properties (appearance, flavour, texture), functional properties (how an ingredient behaves in use), convenience and shelf life, and integrity (authenticity and honest description). Garvin's eight dimensions of quality, performance, features, reliability, conformance, durability, serviceability, aesthetics and perceived quality, remind managers that customers judge more than conformance to a specification.
Several terms describe how quality is managed. Quality control (QC) is the operational activity of measuring and checking products and processes against specifications and acting on deviations. Quality assurance (QA) is the planned and systematic set of activities, including procedures, training, supplier approval, audits and documentation, that gives confidence that quality requirements will be met; it prevents problems instead of only detecting them. Quality management covers all coordinated activities to direct and control an organisation with regard to quality, including policy, objectives and improvement. Total quality management (TQM) extends this to every function and every person, with customer focus and continual improvement as organising principles. A specification sets the characteristics required, with a target value and tolerance limits; specifications are written for raw materials, packaging, intermediates and finished products, and agreed with suppliers and customers.
Modern quality management grew from the work of a small number of pioneers. Walter Shewhart introduced control charts in the 1920s and distinguished common-cause from special-cause variation. W. Edwards Deming taught statistical methods and management in post-war Japan and summarised his philosophy in fourteen points, including ending dependence on mass inspection, improving constantly and forever, and driving out fear; he popularised the plan-do-check-act (PDCA) cycle, also called the Shewhart or Deming cycle. Joseph Juran described the quality trilogy of planning, control and improvement and emphasised the vital few problems that cause most losses (the Pareto principle). Philip Crosby promoted zero defects and the idea that quality is free, because the cost of preventing failures is less than the cost of failures. Kaoru Ishikawa developed the cause-and-effect diagram and quality circles, and Armand Feigenbaum the idea of total quality control.
The cost of quality is usually analysed in four categories (the prevention-appraisal-failure model). Prevention costs are spent to stop defects happening: training, supplier development, process design, maintenance and quality planning. Appraisal costs are spent to find defects: inspection, testing, calibration and audits. Internal failure costs arise from defects found before delivery: scrap, rework, downgrading, re-inspection and downtime. External failure costs arise after delivery: complaints, returns, recalls, liability, lost customers and damage to reputation; in food they can be very large, because a single recall can cost more than years of prevention. The general lesson is that spending on prevention usually reduces total cost by reducing failure costs. Quality systems in food businesses are now often integrated: a single management system may meet ISO 9001 for quality, ISO 22000 or a certification scheme recognised by the Global Food Safety Initiative for safety, and environmental and occupational standards, with common documentation, audits and management review.
Ishikawa's seven basic tools let any team collect and analyse data on quality problems. A check sheet is a structured form for recording how often defects of each type occur, where and when. A histogram shows the distribution of a measured variable, revealing its centre, spread, shape and whether it fits within specification limits. A Pareto chart ranks causes or defect types by frequency or cost, as bars in descending order with a cumulative line, showing the vital few that account for most of the problem, often about 80 per cent of the effect from about 20 per cent of the causes. A cause-and-effect diagram (fishbone or Ishikawa diagram) organises possible causes of an effect under headings, commonly the six Ms: materials, methods, machines, manpower (people), measurement and mother nature (environment). A scatter diagram plots two variables against each other to suggest a relationship, such as oven temperature against moisture of biscuits, although correlation does not prove causation. A control chart plots a process measure over time against statistically calculated limits (chapter 4). The seventh tool is either stratification, separating data by source (line, shift, supplier) to reveal differences, or a flow chart of the process; lists differ between authors.
Seven management and planning tools support the earlier, less numerical stages of improvement: affinity diagrams, interrelationship digraphs, tree diagrams, matrix diagrams, prioritisation matrices, process decision programme charts and activity network diagrams. Brainstorming and the five whys, asking 'why?' repeatedly until a root cause is reached, are simple techniques for root-cause analysis, used in corrective action after complaints, nonconformities and audit findings.
Kaizen, the Japanese term for continuous improvement, emphasises many small improvements made by the people who do the work. 5S organises the workplace: sort (remove what is not needed), set in order (a place for everything), shine (clean and inspect), standardise (make the first three routine) and sustain (keep it up); in food plants it overlaps with hygiene and foreign-body control. Poka-yoke, error-proofing, designs processes so that mistakes cannot happen or are detected immediately, such as a label printer that will not print without a scanned batch code. Quality circles are small groups of workers who meet regularly to solve problems in their area.
Lean manufacturing, derived from the Toyota production system, aims to eliminate waste (muda), classically in seven forms: overproduction, waiting, transport, over-processing, inventory, motion and defects. Six Sigma, developed in electronics manufacturing, aims to reduce variation so that defects are very rare; a process at six-sigma capability, allowing for a long-term drift of 1.5 standard deviations, produces about 3.4 defects per million opportunities. Its projects follow the DMAIC cycle: define, measure, analyse, improve, control. Many companies combine both as Lean Six Sigma.
Failure mode and effects analysis (FMEA) is a structured way to anticipate failures. For each step of a process or each component of a design, the team lists possible failure modes, their effects and causes, and rates severity, occurrence and detection, each typically on a 1 to 10 scale; the risk priority number (RPN) is the product of the three, from 1 to 1000. Failure modes with high RPN, or with high severity whatever the RPN, are given action first. For example, a failure with severity 8, occurrence 3 and detection 5 has RPN 120. FMEA is used in packaging line design, equipment selection and new product development, and it parallels the hazard analysis of HACCP, which focuses specifically on food-safety hazards. Quality function deployment (QFD), with its house of quality matrix, translates customer needs into technical requirements during product design, and benchmarking compares performance and practices with the best in the industry.
ISO 9001 is the international standard that specifies requirements for a quality management system; it is generic, applying to any organisation, and the current edition is ISO 9001:2015 (with a 2024 amendment adding consideration of climate change to the organisation's context). It rests on the seven quality management principles set out in ISO 9000: customer focus, leadership, engagement of people, the process approach, improvement, evidence-based decision making, and relationship management. The standard uses the high-level structure common to ISO management-system standards, which makes it easy to integrate with ISO 22000 (food safety), ISO 14001 (environment) and ISO 45001 (occupational health and safety). Its ten clauses are: 1 scope, 2 normative references, 3 terms and definitions, 4 context of the organisation (internal and external issues, interested parties, scope of the system, processes), 5 leadership (commitment, quality policy, roles and responsibilities), 6 planning (actions to address risks and opportunities, quality objectives, planning changes), 7 support (resources, including monitoring and measuring resources and organisational knowledge; competence; awareness; communication; documented information), 8 operation (planning, customer requirements, design and development, control of externally provided processes and products, production and service provision, release, control of nonconforming outputs), 9 performance evaluation (monitoring and measurement, customer satisfaction, internal audit, management review) and 10 improvement (nonconformity and corrective action, continual improvement).
Two features of the 2015 edition are often examined. Risk-based thinking replaced the earlier separate requirement for preventive action: the organisation must determine risks and opportunities and plan actions to address them throughout the system. The documentation requirements were simplified into 'documented information', which the organisation must either maintain (documents such as procedures and specifications) or retain (records as evidence of results); a quality manual is no longer mandatory. The process approach, linked to the PDCA cycle, asks the organisation to manage its activities as interrelated processes with defined inputs, outputs, owners and performance measures.
Audits check whether a system meets requirements and is effective. First-party audits are internal audits by the organisation itself; second-party audits are carried out by a customer on a supplier; third-party audits are carried out by an independent certification body. ISO 19011 gives guidance on auditing management systems, including audit programmes, auditor competence and the conduct of audits. Third-party certification to ISO 9001 is issued by certification bodies that are themselves accredited to ISO/IEC 17021-1 by a national accreditation body; a typical cycle is an initial two-stage certification audit, surveillance audits at least annually, and recertification every three years. Certification shows that a system conforms to the standard, not that every product is good; ISO 9001 is not a food-safety standard and does not replace HACCP-based food safety management.
Nonconformities found by audits, inspections, complaints or monitoring are handled in two steps. Correction deals with the immediate problem, for example segregating and reworking a nonconforming batch. Corrective action eliminates the cause so that the problem does not recur: investigating the root cause, implementing changes, and verifying that they worked. A common audit finding is that organisations take correction without corrective action, so the same problem returns. Management review, at planned intervals, has top management examine audit results, customer feedback, process performance, the status of actions and changes, and decide on improvements and resources. ISO 9001 has been widely adopted by food manufacturers, ingredient suppliers, packaging makers and laboratories (alongside ISO/IEC 17025), but food retailers and brand owners usually also require certification to a food-safety scheme. Integrated management systems avoid duplication by sharing document control, internal audit, corrective action and management review across standards.
Every process varies. Shewhart distinguished common-cause variation, the many small, random sources inherent in a stable process, from special-cause (assignable) variation, arising from identifiable events such as a worn nozzle, a new batch of raw material or an untrained operator. A process with only common-cause variation is in statistical control and predictable; a control chart is a tool for detecting special causes so that they can be found and removed, and for avoiding over-adjustment (tampering) of a process that is in control, which increases variation. Control limits are calculated from the process data, conventionally at three standard deviations of the plotted statistic either side of the centre line; they describe what the process does, and they are not the same as specification limits, which describe what the customer requires.
For measured variables, such as fill weight, moisture, pH or seal strength, small rational subgroups (often 4 or 5 consecutive units) are taken at intervals. The X̄ (mean) chart tracks the process centre and the R (range) chart its short-term spread. With the grand mean X̄̄ and the mean range R̄ from at least 20 to 25 subgroups, the limits are UCL = X̄̄ + A₂R̄ and LCL = X̄̄ − A₂R̄ for the X̄ chart, and UCL = D₄R̄ and LCL = D₃R̄ for the R chart. For subgroups of 5, A₂ = 0.577, D₃ = 0 and D₄ = 2.114; for subgroups of 4, A₂ = 0.729, D₃ = 0 and D₄ = 2.282. For example, with X̄̄ = 500.0 g, R̄ = 4.0 g and n = 5, the X̄ chart limits are 500.0 ± 0.577 × 4.0, that is 497.7 to 502.3 g, and the R chart upper limit is 8.5 g. The process standard deviation can be estimated as R̄/d₂, with d₂ = 2.326 for n = 5. When subgroups are larger, X̄ and s charts are used; when only one measurement is available per period, as with a batch analysis, an individuals and moving range chart is used, with limits at X̄ ± 2.66 × MR̄.
Attribute data count defectives or defects. The p chart plots the proportion defective in samples of size n, with limits p̄ ± 3√(p̄(1 − p̄)/n); the np chart plots the number defective in samples of constant size. The c chart plots the number of defects per inspection unit (such as blemishes per carton or foreign particles per sample) with limits c̄ ± 3√c̄, and the u chart plots defects per unit when the sample size varies. A negative lower limit is set to zero. For example, if 2 per cent of jars are rejected for poor seals in samples of 200, the p chart limits are 0.02 ± 3√(0.02 × 0.98/200) = 0.02 ± 0.0297, giving an upper limit of about 0.050 and a lower limit of zero.
A point outside the control limits is the clearest signal of a special cause, but patterns inside the limits also matter. The Western Electric rules, widely used in industry, flag one point beyond 3 sigma; two of three consecutive points beyond 2 sigma on the same side; four of five consecutive points beyond 1 sigma on the same side; and eight consecutive points on one side of the centre line (some sets of rules use runs of seven or nine and add trends of six or seven points rising or falling). Each extra rule increases sensitivity but also the rate of false alarms. For detecting small, sustained shifts, cumulative sum (CUSUM) and exponentially weighted moving average (EWMA) charts are more sensitive than Shewhart charts. Good practice is to choose subgroups so that variation within them reflects only common causes, to set limits from a period when the process was stable, to recalculate limits only after deliberate process changes, and to record on the chart the actions taken, so that the chart becomes a history of the process. Control charts are applied in food plants to net contents, moisture, salt, fat, pH, temperatures, seal strengths and defect counts, and in laboratories to the results of control samples (chapter 9).
A process can be in statistical control and still unable to meet its specification, if its natural spread is too wide or its centre is in the wrong place. Process capability indices compare the spread of the process with the width of the specification. The potential capability index Cp = (USL − LSL)/(6σ) compares the specification width (between the upper and lower specification limits) with the six-sigma spread of the process, ignoring where the process is centred. The actual capability index Cpk = min[(USL − μ)/(3σ), (μ − LSL)/(3σ)] takes centring into account by measuring the distance from the mean to the nearer specification limit in units of three standard deviations. Cpk equals Cp only when the process is centred midway between the limits, and Cpk is always less than or equal to Cp. For a one-sided specification, such as a minimum seal strength or a maximum moisture, only the relevant half is calculated (Cpl or Cpu).
For example, if pack moisture must lie between 3.0 and 5.0 per cent, the process mean is 4.2 per cent and σ is 0.25 per cent, then Cp = 2.0/1.5 = 1.33 and Cpk = min[(5.0 − 4.2)/0.75, (4.2 − 3.0)/0.75] = min[1.07, 1.60] = 1.07; the process could be capable but is off-centre, and re-centring at 4.0 per cent would raise Cpk to 1.33. A Cp of 1.0 means the process spread exactly fills the specification, and, if centred and normally distributed, about 0.27 per cent (2700 parts per million) of output falls outside; Cpk = 1.0 means about 0.135 per cent falls outside on the nearer side. Many industries require a minimum Cpk of 1.33 for established processes and 1.67 for critical characteristics or new processes; a six-sigma process has Cp = 2.0.
Capability indices are meaningful only under conditions that are often overlooked. The process must be in statistical control, because a capability figure computed from an unstable process predicts nothing; the data should be approximately normally distributed, or a suitable transformation or a non-normal method used, because many food characteristics, such as microbial counts and impurity levels, are skewed; the measurement system must be adequate, with measurement variation small compared with the tolerance (a gauge repeatability and reproducibility study assesses this); and enough data are needed, as capability estimates from small samples are very uncertain.
Two estimates of σ give two families of indices. Cp and Cpk usually use the within-subgroup estimate R̄/d₂ (or s̄/c₄), describing short-term capability; Pp and Ppk, the performance indices, use the overall standard deviation of all data, which includes between-subgroup drift, describing long-term performance. A large gap between Cpk and Ppk shows that the process drifts between subgroups. Capability studies are used to decide whether a new line or supplier can meet a specification, to set realistic specifications and target settings, to justify reduced inspection when capability is high, and to prioritise improvement. In the food industry, a common application is to fill-weight control, where a process with low capability must be run with a higher target to keep underweight packs within the legal limits, giving away product; improving capability by reducing variation allows the target to be lowered safely, saving material. For instance, if a filler with σ = 2.0 g must keep the mean at least three standard deviations above a 500 g minimum, the target is 506 g; halving σ to 1.0 g allows a target of 503 g.
Acceptance sampling decides whether to accept or reject a lot on the basis of a sample, when inspecting every unit is impossible (destructive tests), too costly or less reliable because of inspector fatigue. It does not control or improve quality; it gives a decision rule with known risks. Plans are by attributes, counting defective units, or by variables, using measured values and their distribution, which needs smaller samples for the same protection but assumes a known distribution and applies to one characteristic at a time. A single sampling plan by attributes is defined by the sample size n and the acceptance number c: inspect n units; accept the lot if the number of defectives d ≤ c, reject if d > c. Double, multiple and sequential plans take further samples when the first result is borderline and can reduce average sample size.
The performance of a plan is described by its operating characteristic (OC) curve, which plots the probability of accepting a lot, Pa, against the true fraction defective p in the lot. For large lots, Pa is given by the binomial distribution, Pa = Σ (from d = 0 to c) of C(n, d)pᵈ(1 − p)ⁿ⁻ᵈ, or approximately by the Poisson distribution with mean np. For a plan with n = 20 and c = 0, a lot with 5 per cent defectives is accepted with probability 0.95²⁰ = 0.36, and a lot with 1 per cent defectives with probability 0.82. For n = 50 and c = 1, a lot with 2 per cent defectives (np = 1) is accepted with probability about e⁻¹(1 + 1) = 0.74. An ideal OC curve would accept all lots below a chosen quality and reject all above it, a vertical step; real curves slope, and they become steeper (more discriminating) as n increases. Increasing c with n kept constant makes the plan more lenient.
Two risks are defined on the OC curve. The producer's risk α is the probability of rejecting a lot of good quality, conventionally set at about 5 per cent at the acceptable quality level; the consumer's risk β is the probability of accepting a lot of poor quality, often set at about 10 per cent at the limiting quality (also called the lot tolerance percent defective). The acceptable quality limit (AQL), in the international standard ISO 2859-1, is the worst tolerable process average when a continuing series of lots is submitted; it is not a level of defects that is desirable or acceptable to the consumer in any single lot. With rectifying inspection, in which rejected lots are screened and defectives replaced, the average outgoing quality (AOQ ≈ p × Pa) rises and then falls as incoming quality worsens; its maximum is the average outgoing quality limit (AOQL).
ISO 2859-1, derived from the former United States military standard MIL-STD-105, is the most widely used attribute sampling system. The user chooses an inspection level (general levels I, II and III, with II the default, and special levels S-1 to S-4 for small samples) and the lot size, which give a sample-size code letter; the code letter and the AQL give n and the acceptance and rejection numbers. The system includes switching rules between normal, tightened and reduced inspection based on recent lot history, and it is meant to be used for a continuing series of lots, not isolated ones. ISO 3951 gives the corresponding variables plans. For food in international trade, the Codex General Guidelines on Sampling (CXG 50-2004) explain sampling plans, the AQL and limiting quality concepts and their use for compositional and contaminant provisions in Codex standards, while specific Codex standards set plans for particular commodities. Sampling plans also depend on drawing a representative, random sample from the lot; a statistically correct plan applied to a sample taken from one corner of a pallet gives a false answer.
Microbiological criteria for foods use sampling plans developed by the International Commission on Microbiological Specifications for Foods (ICMSF) and adopted in Codex and in many national regulations. A two-class plan divides results into acceptable and unacceptable using one limit, m. It is defined by n, the number of sample units tested from the lot; c, the maximum number of units allowed to exceed m; and m itself. For pathogens such as Salmonella, plans are usually two-class with c = 0 and m expressed as absence in a stated analytical unit, for example 'not detected in 25 g' in each of n units; one positive unit means the lot is rejected. A three-class plan uses two limits: m separates good quality from marginal quality, and M separates marginal from unacceptable. Units with counts at or below m are acceptable; units above m but not above M are marginal; any unit above M makes the lot unacceptable; and the lot is also rejected if more than c units are marginal. Three-class plans are typical for indicator organisms and for counts of organisms such as Escherichia coli, Staphylococcus aureus, Bacillus cereus or aerobic colonies, where a few marginal units are tolerable but none may be grossly contaminated.
For example, consider a three-class plan with n = 5, c = 2, m = 10² and M = 10³ cfu/g. Results of 50, 300, 800, 90 and 40 cfu/g give two marginal units (300 and 800) and none above M, so the lot is accepted; results of 50, 300, 800, 600 and 40 give three marginal units, more than c, and the lot is rejected; and a single result of 2000 cfu/g rejects the lot whatever the others are. The stringency of a plan is chosen from the severity of the hazard and from whether conditions of handling after sampling are expected to reduce, not change or increase the hazard; the ICMSF arranged these combinations into fifteen 'cases', with more units and smaller c for more severe hazards and conditions that allow growth.
Microbiological sampling plans have statistical limits that every food technologist should understand. Microorganisms are usually unevenly distributed in a lot, and concentrations tend to follow a log-normal distribution, so a few units may be heavily contaminated while most are clean. A plan with c = 0 accepts a lot when all n units test negative, and its probability of acceptance at a true proportion p of contaminated units is (1 − p)ⁿ. With n = 5, a lot in which 10 per cent of units are contaminated is still accepted with probability 0.9⁵ = 0.59; with n = 30, that probability falls to about 0.04; and to have 95 per cent confidence of detecting a lot in which 5 per cent of units are contaminated, about 59 units are needed, because 0.95⁵⁹ ≈ 0.048. Testing therefore cannot prove that a lot is free of a pathogen; it can only detect contamination above a certain prevalence with a stated confidence.
For this reason, Codex principles for the establishment and application of microbiological criteria (CXG 21-1997) emphasise that end-product testing is a verification tool and that safety is built into food by good hygiene practices and HACCP. Microbiological criteria are of three types: food safety criteria, which define the acceptability of a product or lot placed on the market (such as limits for Listeria monocytogenes in ready-to-eat foods or Salmonella in specified foods); process hygiene criteria, which indicate the acceptable functioning of a production process and trigger corrective action when exceeded; and guidelines or specifications agreed between businesses. A criterion states the food to which it applies, the microorganism or toxin, the analytical method, the sampling plan, the limits, the stage of the chain where it applies and the action to be taken when it is not met. Results are compared with the limits taking the method into account: the analytical unit (for example 25 g) and the reference method are part of the criterion, and composite testing of several units in one enrichment is allowed in some cases when it does not reduce sensitivity.
Quality control is organised at three points. Raw-material control begins with supplier approval: assessing suppliers by questionnaire, audit and certification status, agreeing specifications, and monitoring their performance. Each delivery is checked on receipt against the specification and the risk: vehicle condition and temperature, packaging and labelling, identity, sensory attributes, and key analyses such as moisture, fat, acidity, pesticide residues or mycotoxins in grains and nuts, or antibiotic residues in milk, sometimes relying on the supplier's certificate of analysis with periodic verification testing by the receiving plant. Nonconforming deliveries are rejected, held or accepted under concession with records.
In-process control monitors the quality parameters that determine the final product at the stage where they can still be corrected: dough temperature and mixing time, Brix and pH of a jam before filling, fill weight and headspace, seal integrity, can double-seam dimensions by tear-down examination and seam projection, oven temperatures and product colour, metal detection and checkweighing. Quality control points differ from HACCP critical control points: CCPs control food-safety hazards with critical limits, whereas quality control points govern quality characteristics, although many plants monitor both with the same systems. Statistical process control (chapter 4) is most effective here.
Finished-product control confirms that the product meets its specification before and after release: sensory evaluation against a standard, chemical and physical analysis, microbiological testing as verification, packaging and label checks (including allergen statements, date coding and legibility), and net-quantity control. Many countries base net-quantity rules for prepackages on OIML Recommendation R 87, the average system: the average actual quantity of a batch must not be less than the nominal quantity; only a small proportion of packages may fall short by more than the tolerable negative error, a quantity that depends on the nominal size (for example 15 g for packages between 500 g and 1 kg); and no package may fall short by more than twice that error. Retained samples, positive release where required, control of nonconforming product, traceability records, and the handling of customer complaints complete the system, and complaint trends feed back into corrective action.
Non-destructive and online methods measure quality without destroying the product, often on every unit at line speed, and they are replacing sampling-based inspection for many characteristics. Machine vision systems use cameras and image analysis to grade fruit and vegetables by colour, size, shape and defects, check fill levels, cap position and label placement, read date codes, and inspect baked goods; optical sorters eject discoloured grains, nuts and pulses at high throughput. Near-infrared (NIR) spectroscopy measures moisture, fat, protein and sugar in seconds, at-line or online on conveyors and in pipes, using chemometric calibrations against reference methods; it is standard for grain and flour, dairy powders, meat and feed. Hyperspectral imaging combines imaging and spectroscopy to map composition and detect contamination or bruising. Checkweighers weigh every pack and reject under- and overweight ones, feeding data to filler control.
Foreign-body detection protects consumers from physical hazards. Metal detectors detect ferrous, non-ferrous and stainless-steel contaminants; stainless steel is generally the hardest to detect, and wet, salty or conductive products cause a 'product effect' that reduces sensitivity. Detectors are verified at set intervals with certified test pieces of each metal type passed through the aperture in the product, and the reject mechanism is checked too; test-piece sizes are set by risk assessment and customer codes. X-ray inspection detects dense contaminants such as metal, glass, stone, bone and some hard plastics, and also checks fill level, missing items and seal contamination. Other non-destructive techniques include ultrasound for texture and fat content, acoustic and impact methods for fruit firmness, time-domain nuclear magnetic resonance for solid fat content and moisture, and electronic noses for aroma. Any online method needs calibration against a reference, validation for the product range, routine verification, and procedures for what happens to product when the instrument fails, because an instrument that is assumed to work but has drifted is worse than no instrument at all.
A laboratory result is the basis of decisions to release or reject product, to enforce the law and to accept goods in trade, so laboratories need their own quality system. ISO/IEC 17025:2017, General requirements for the competence of testing and calibration laboratories, is the international standard for this. It is organised in five main clauses. General requirements (clause 4) cover impartiality, including identifying and managing risks to it, and confidentiality. Structural requirements (clause 5) cover the legal entity, management responsibilities and defined scope of activities. Resource requirements (clause 6) cover competent personnel with records of training, authorisation and monitoring; suitable facilities and environmental conditions (temperature, humidity, separation of incompatible activities such as trace analysis and high-level standards); equipment, calibrated and maintained, with records; metrological traceability of measurement results to the International System of Units through an unbroken chain of calibrations, normally by accredited calibration laboratories or national metrology institutes; and control of externally provided products and services. Process requirements (clause 7) cover review of requests and contracts, selection, verification and validation of methods, sampling, handling of test items, technical records, evaluation of measurement uncertainty, ensuring the validity of results, reporting, complaints, nonconforming work and control of data and information management. Management system requirements (clause 8) can be met either directly (option A) or through a management system conforming to ISO 9001 (option B).
The 2017 revision introduced risk-based thinking in place of a preventive-action clause, aligned the structure with other conformity-assessment standards, and added explicit requirements for decision rules: when a laboratory states whether a result conforms to a specification or limit, it must document and agree the rule it uses to take measurement uncertainty into account (chapter 11).
Clause 7.7 asks laboratories to monitor the validity of their results by planned activities, and food laboratories use a standard set. Internal quality control includes running reagent blanks to detect contamination; replicate analyses to monitor precision; spiked samples or certified reference materials to monitor recovery and trueness; and control samples of a stable, homogeneous material analysed with each batch, whose results are plotted on Shewhart control charts with warning limits at ±2 standard deviations and action limits at ±3 standard deviations; a result beyond an action limit stops reporting until the cause is found. Calibration curves are checked with independent standards, and instrument performance checks (such as system-suitability tests in chromatography) are run before analyses. External quality control comes from proficiency testing and interlaboratory comparisons (chapter 12).
Other practices make results defensible. Sample receipt and chain-of-custody records show who held a sample and when, which is essential for official and legal samples; samples are stored to prevent deterioration, and retained portions are kept for re-testing. Technical records allow the test to be repeated as performed: who did it, on which equipment, with which reagents and standards, and with raw data. Laboratory information management systems (LIMS) handle samples, results and reports with controlled access and audit trails. Reports state the method, results with units and, where relevant, measurement uncertainty and any statement of conformity. Good laboratory practice (GLP), as set out in the OECD principles, is a separate quality system for non-clinical safety studies submitted to regulators, such as toxicology studies of additives and pesticides; it concerns study organisation and data integrity, not routine testing, and a laboratory may need both GLP compliance and ISO/IEC 17025 accreditation for different work. Microbiology laboratories also follow ISO 7218 on general requirements for food microbiology and ISO 11133 on culture media performance.
A laboratory must show that each method is fit for its intended purpose. A standard method published by a recognised body (ISO, AOAC International, Codex-adopted, national standards) that has been collaboratively studied needs verification in the laboratory: evidence that the laboratory can achieve the published performance with its own staff, equipment and matrices. A non-standard method, a laboratory-developed method, or a standard method used outside its scope or modified needs full validation. Validation establishes performance characteristics. Selectivity (specificity) is the ability to measure the analyte without interference from other components of the matrix. Linearity and working range define the concentration interval over which the response is proportional to concentration, tested with calibration standards at several levels. Trueness is the closeness of the mean of many results to the true or accepted value, expressed as bias, and estimated with certified reference materials, comparison with a reference method, or recovery of spiked analyte. Precision is the closeness of agreement between independent results and is expressed as a standard deviation or relative standard deviation under stated conditions: repeatability (same method, laboratory, analyst, equipment, short interval), intermediate precision (within one laboratory over days, analysts and equipment) and reproducibility (between laboratories, from collaborative studies). Accuracy combines trueness and precision. Robustness (ruggedness) shows how small deliberate changes in method parameters, such as extraction time or pH, affect results. Measurement uncertainty completes the picture (chapter 11).
Recovery is calculated as R% = (C_spiked − C_unspiked)/C_added × 100. For example, if a sample containing 2.0 mg/kg of an analyte is spiked with 5.0 mg/kg and the spiked sample gives 6.6 mg/kg, the recovery is (6.6 − 2.0)/5.0 × 100 = 92 per cent. Acceptable recovery ranges depend on concentration and purpose and are given in validation guidelines; wider ranges are accepted at trace levels.
The limit of detection (LOD) is the lowest concentration that can be distinguished reliably from a blank, and the limit of quantification (LOQ) the lowest concentration that can be measured with acceptable precision and trueness. They are commonly estimated from the standard deviation of blank or low-level results: LOD as the mean blank plus 3 standard deviations (or 3.3σ/slope of the calibration curve), and LOQ as the mean blank plus 10 standard deviations (or 10σ/slope). For example, if ten blank results have a standard deviation of 0.02 µg/kg in concentration terms and a mean of zero, the LOD is about 0.06 µg/kg and the LOQ about 0.2 µg/kg. Results between the LOD and LOQ are reported as detected but not quantified, or with qualification. For enforcement, the LOQ must be well below the legal limit being checked.
The repeatability limit r = 2.8 × s_r is the value below which the absolute difference between two results obtained under repeatability conditions is expected to lie with 95 per cent probability (2.8 ≈ 1.96 × √2), and the reproducibility limit R = 2.8 × s_R does the same for results from different laboratories; the precision standard ISO 5725 and many official methods give these values. If duplicates differ by more than r, the analysis is repeated. Collaborative studies of food methods showed that reproducibility depends mainly on concentration, which Horwitz summarised in an empirical equation: the predicted relative standard deviation of reproducibility is RSD_R (%) = 2^(1 − 0.5 log₁₀C), where C is the concentration as a mass fraction. At 1 per cent (C = 0.01) this gives 4 per cent, at 1 mg/kg (C = 10⁻⁶) 16 per cent, and at 1 µg/kg (C = 10⁻⁹) about 45 per cent; at very low concentrations a modified value (about 22 per cent below roughly 120 µg/kg) is often used instead. The HorRat ratio, observed RSD_R divided by predicted RSD_R, is used to judge collaborative studies, with values of roughly 0.5 to 2 regarded as acceptable. Microbiological methods are validated differently, with qualitative characteristics such as sensitivity, specificity, false-positive and false-negative rates and the level of detection, following the ISO 16140 series for validating alternative methods against reference methods.
Every measurement result is an estimate, and its measurement uncertainty is a parameter that characterises the dispersion of values that could reasonably be attributed to the measurand. The international framework is the Guide to the Expression of Uncertainty in Measurement (GUM, JCGM 100:2008), with practical guides for chemical analysis from Eurachem and CITAC and for microbiology in ISO 19036. In the bottom-up (modelling) approach, the analyst writes the equation by which the result is calculated, identifies every source of uncertainty (sampling and sub-sampling, weighing, volumetric operations, purity of standards, calibration, recovery, repeatability, instrument effects), quantifies each as a standard uncertainty u, and combines them.
Standard uncertainties are evaluated in two ways. Type A evaluation uses statistics of repeated observations: the standard uncertainty of a mean of n results is s/√n. Type B evaluation uses other information, such as calibration certificates, manufacturers' tolerances, published data or judgement. A tolerance of ±a with no other information is treated as a rectangular distribution with standard uncertainty a/√3; if values near the centre are more likely, a triangular distribution with a/√6 is used; a certificate that gives an expanded uncertainty with a coverage factor k is divided by k. For example, a 100 mL volumetric flask with a tolerance of ±0.08 mL has a standard uncertainty from the tolerance of 0.08/√3 = 0.046 mL.
Standard uncertainties are combined by the law of propagation of uncertainty. For a result that is a sum or difference, the absolute standard uncertainties are combined in quadrature: u_c = √(u₁² + u₂² + …). For a result that is a product or quotient, the relative standard uncertainties are combined in quadrature: u_c/y = √((u₁/x₁)² + (u₂/x₂)² + …). For example, relative standard uncertainties of 1 per cent (weighing and volume), 2 per cent (calibration) and 2 per cent (repeatability) give a combined relative standard uncertainty of √(1 + 4 + 4) = 3 per cent. The largest components dominate, so effort to reduce uncertainty is best spent on them. The top-down approach instead estimates uncertainty from method validation and quality-control data, combining long-term within-laboratory reproducibility with an estimate of bias uncertainty from certified reference materials or proficiency testing, as in the Nordtest approach; it is widely used in food laboratories because it captures effects that a model might miss.
The combined standard uncertainty corresponds to one standard deviation. To give an interval that contains a large fraction of the values that could reasonably be attributed to the measurand, it is multiplied by a coverage factor k to give the expanded uncertainty U = k·u_c. A coverage factor of 2 gives a level of confidence of approximately 95 per cent when the distribution is approximately normal and the degrees of freedom are adequate. A result is reported as, for example, 12.4 ± 0.8 mg/kg, where the reported uncertainty is an expanded uncertainty calculated with a coverage factor of 2, giving a level of confidence of approximately 95 per cent. Uncertainty is usually given to one or two significant figures, and the result rounded to match.
Uncertainty matters most when a result is close to a legal limit or specification. Several decision rules are in use, and ISO/IEC 17025 requires the laboratory to state which one it applies. In enforcement of maximum limits for contaminants and residues, many authorities, following Codex guidance on measurement uncertainty, consider a sample non-compliant only if the result minus its expanded uncertainty exceeds the limit, so that the decision is made 'beyond reasonable doubt' with respect to measurement. For example, with a maximum limit of 10 µg/kg and an expanded uncertainty of 2.0 µg/kg, a result of 11.5 µg/kg (lower bound 9.5) would not be declared non-compliant under that rule, whereas 12.5 µg/kg (lower bound 10.5) would. Food businesses checking their own products before release often apply the opposite guard band, releasing only if the result plus U is below the limit. The same result can therefore lead to different decisions depending on who decides and with which rule, which is why decision rules are agreed before testing. Results are also corrected for recovery, or not, according to the rules for the analysis; some regulations for mycotoxins and residues specify whether and how recovery correction is applied, and the uncertainty of the recovery correction is then part of the budget.
Proficiency testing (PT) is the evaluation of a laboratory's performance against pre-established criteria by means of interlaboratory comparisons. A PT provider, ideally accredited to ISO/IEC 17043, distributes homogeneous, stable portions of a test material, such as milk powder with a known level of aflatoxin M1 or a meat sample spiked with Salmonella, to many laboratories, which analyse them by their routine methods and report results. Performance is commonly expressed as a z-score, z = (x − X)/σ_pt, where x is the laboratory's result, X the assigned value (from formulation, a reference laboratory or the consensus of participants) and σ_pt the standard deviation for proficiency assessment, often derived from the Horwitz equation or from fitness-for-purpose criteria. Under ISO 13528, |z| ≤ 2 is satisfactory, 2 < |z| < 3 questionable, and |z| ≥ 3 unsatisfactory, requiring investigation and corrective action. For example, a result of 4.6 mg/kg against an assigned value of 4.0 mg/kg with σ_pt = 0.4 mg/kg gives z = 1.5, satisfactory; 5.4 mg/kg would give z = 3.5, unsatisfactory. When participants report uncertainties, the En number, En = (x − X)/√(U_lab² + U_ref²), is used, with |En| ≤ 1 satisfactory. Accreditation bodies require participation in PT where schemes exist, at a frequency suited to the scope, and review the results and the actions taken.
Reference materials provide the link between a laboratory's measurements and accepted values. A reference material is sufficiently homogeneous and stable with respect to specified properties; a certified reference material (CRM) is one characterised by a metrologically valid procedure, with a certificate giving the certified value, its uncertainty and a statement of metrological traceability. Producers of reference materials are accredited to ISO 17034, and ISO Guide 35 describes characterisation and certification. Matrix CRMs, such as a milk powder certified for its mineral content or a cereal certified for mycotoxins, are used to check trueness of whole methods; pure-substance CRMs are used to prepare calibration standards. A bias found with a CRM is assessed against the combined uncertainty of the result and the certified value. Because CRMs are costly and limited in range, laboratories also prepare in-house reference or control materials, characterised against CRMs or by repeated analysis, for routine control charts.
Accreditation is third-party attestation that a conformity assessment body, such as a laboratory, certification body or inspection body, is competent to carry out specific tasks. It differs from certification: an ISO 9001 certificate says that a management system conforms to a standard, while ISO/IEC 17025 accreditation says that a laboratory is technically competent to produce valid results for the tests listed in its scope. Accreditation bodies operate to ISO/IEC 17011. Most countries have a single national accreditation body, often designated by government; examples from different regions include bodies in the United Kingdom, Germany, the United States (where several bodies operate), India, China, Japan, Australia, South Africa, Brazil and Canada.
International recognition is organised in two networks. The International Laboratory Accreditation Cooperation (ILAC) covers laboratories (testing, calibration, medical) and inspection bodies, and operates a Mutual Recognition Arrangement among signatory accreditation bodies; the International Accreditation Forum (IAF) covers certification bodies through its Multilateral Recognition Arrangement. These are linked to regional cooperations, such as European co-operation for Accreditation, the Asia Pacific Accreditation Cooperation, the African Accreditation Cooperation, the Inter-American Accreditation Cooperation and the Arab Accreditation Cooperation, which carry out peer evaluations of their members. A signatory's accredited results are accepted by the other signatories as equivalent, the principle summarised as 'tested once, accepted everywhere', which reduces technical barriers to trade.
The accreditation process typically includes application and definition of the scope (the specific tests, methods, matrices and ranges); review of documentation; an on-site assessment by a lead assessor and technical experts who examine records, interview staff and witness tests; reporting of nonconformities, which the laboratory must close with corrective actions and evidence; an accreditation decision; and then regular surveillance (often annual) and reassessment at the end of each cycle, the length of which varies by body. Regulators rely on this system: many jurisdictions require official control laboratories to be accredited to ISO/IEC 17025 for the methods they use, and the Codex guidelines for the assessment of the competence of testing laboratories involved in the import and export control of foods (CXG 27-1997) recommend accreditation, participation in proficiency testing, validated methods and internal quality control for laboratories whose results are used in trade.
A student uses this subject to analyse laboratory and pilot-plant data properly: drawing X̄-R charts of fill weights, calculating Cp and Cpk, constructing an OC curve for a plan, interpreting a three-class microbiological result, calculating recovery, LOD and LOQ in an analytical practical, and writing a simple uncertainty budget. In examinations these calculations, the AQL definition, ISO 9001 clauses and ISO/IEC 17025 requirements recur often.
Five years into a QA or laboratory job, a technologist runs SPC on lines and challenges tampering, sets fill targets from capability, writes raw-material and finished-product specifications and sampling plans, handles nonconformities with root-cause analysis, validates and verifies laboratory methods, maintains control charts and PT records, estimates uncertainty, and prepares the laboratory for accreditation assessments and the plant for certification audits.
A head of quality or laboratory director with twenty-five years' experience designs the quality system across sites, decides where to invest in online inspection versus sampling, sets decision rules for releasing product near legal limits, manages relationships with accreditation and certification bodies and regulators, and uses cost-of-quality and capability data to argue for prevention.
A researcher at this level develops and validates new analytical and rapid methods, designs collaborative studies and PT schemes, works on uncertainty for complex and microbiological methods, models sampling plan performance for heterogeneous contamination, and contributes to the standards (ISO, Codex, AOAC) that define how food is measured and judged.
This manual covers quality assurance and laboratory quality at degree and early professional level. It does not reproduce the tables of ISO 2859-1 or control-chart constants beyond common subgroup sizes, nor national rules on net quantity, official sampling or decision rules, which differ by jurisdiction; values such as minimum capability indices, PT criteria and accreditation cycle lengths are typical and vary by industry and body. Food safety management systems and certification schemes are in food-safety-management-systems, analytical techniques in food-analysis and advanced-analytical-omics, and statistical theory in statistics-experimental-design-food.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.