Dairy Science & Technology · Quality, Regulation & Dairy Business · also Quality Assurance in Dairying, Statistical Quality Control, Total Quality Management
Quality systems, statistical process control, sampling inspection and running an accredited dairy laboratory.
In the exams: ICAR AICE subject 34 Unit 9, subject 35b Unit 1 and subject 36 Unit 6
Quality assurance in a dairy is the set of plans, checks and records that let a plant say, with evidence, that every pouch of milk, block of butter and bag of powder meets what was promised: its composition, its safety, its sensory character and its shelf life. Milk is a perishable, variable raw material collected from thousands of farms, processed within hours and sold under legal standards for fat, solids-not-fat and microbial counts. That makes dairying one of the most measurement-heavy food industries. A plant tests every tanker at reception, monitors every pasteuriser continuously, weighs packs on the filling line and runs a laboratory that may analyse several hundred samples a day for payment, process control and release.
This manual covers four linked areas. The first is the management of quality: what quality, quality control and quality assurance mean, the ideas of the quality pioneers, ISO 9001 and total quality management. The second is statistics for the shop floor: the basic quality tools, control charts for variables and attributes, process capability, acceptance sampling, and Six Sigma and lean methods applied to dairy lines. The third is the laboratory: how to lay out and run a safe dairy laboratory, how ISO/IEC 17025 accreditation works, how measurement uncertainty and method validation are handled, and how audits and proficiency testing prove that results can be trusted. The fourth is shelf-life determination for chilled, ambient and dried dairy products.
The general statistical theory is taught in agri-statistics (statistical-quality-control and sampling-theory-basic) and in agribusiness (quality-management-standards and production-operations-management); food safety systems such as HACCP and ISO 22000 are in food-technology (food-safety-management-systems) and in this track's dairy-food-safety-systems; legal standards are in dairy-standards-law-worldwide; and analytical methods are in dairy-analysis-instrumental. This manual gives a short reminder of each general idea and spends its words on dairy application. In India the subject is examined in ICAR AICE-JRF/SRF Dairy Chemistry (subject 34, Unit 9) and Dairy Microbiology (subject 36, Unit 6), in the quality and standards items of ICAR AIEEA-PG group 16, and in Food Analyst and Food Safety Officer recruitment. Elsewhere it underpins management-system auditor qualifications and the certification of milk laboratories.
ISO 9000 defines quality as the degree to which a set of inherent characteristics of an object fulfils requirements. The requirements may be stated by a customer, set by law or implied by normal expectation. For a dairy product the characteristics fall into five groups. Compositional quality covers fat, solids-not-fat, protein, moisture and salt, most of which are fixed by legal standards of identity and by the product specification. Microbiological quality covers total and coliform counts, pathogens and spoilage organisms. Safety quality covers chemical hazards such as veterinary drug residues, aflatoxin M1, pesticides and adulterants. Sensory quality covers flavour, body and texture, colour and appearance, judged by trained panels against score cards. Functional quality covers how the product performs in use: heat stability of milk for UHT or evaporated milk, solubility and wettability of powder, meltdown of ice cream, stretch of pizza cheese.
David Garvin described eight dimensions of quality: performance, features, reliability, conformance, durability, serviceability, aesthetics and perceived quality. In dairying, conformance (meeting specification) and perceived quality (brand trust) usually dominate, but durability appears as shelf life and aesthetics as appearance and packaging. Juran's phrase "fitness for use" and Crosby's "conformance to requirements" are the two classic short definitions; the first looks outward to the user, the second inward to the specification.
Quality control (QC) is the part of quality management focused on fulfilling quality requirements: the operational techniques of inspection, testing and adjustment that detect nonconforming product. Quality assurance (QA) is the part focused on providing confidence that quality requirements will be fulfilled: the planned, documented system that prevents nonconformity and proves control to customers and regulators. QC is product-oriented and largely reactive; QA is process-oriented and preventive. Both sit inside quality management, which also includes quality policy, objectives, planning and improvement.
In a dairy the platform tests at milk reception are classic QC: organoleptic assessment, temperature, clot-on-boiling, the alcohol test, rapid fat and SNF by an analyser, and screening for antibiotic residues and adulterants. A tanker that fails is rejected. Quality assurance is the system around those tests: approved suppliers, farm hygiene programmes, cleaning-in-place validation, calibrated thermometers on the pasteuriser, documented procedures, training records, internal audits and management review. A plant that relies only on end-product testing finds problems after the milk has been processed; a plant with good QA designs them out. HACCP is the food-safety branch of the same preventive idea and is covered in dairy-food-safety-systems.
Walter Shewhart introduced the control chart and the distinction between chance and assignable causes of variation. W. Edwards Deming built on it: most quality problems come from the system, which only management can change, not from workers. His fourteen points include ending dependence on mass inspection, constant improvement of the system and driving out fear; his Plan-Do-Check-Act (or Plan-Do-Study-Act) cycle is the basis of continual improvement in ISO standards. Joseph Juran described the quality trilogy of planning, control and improvement and popularised the Pareto principle of the vital few and the trivial many. Philip Crosby argued that "quality is free" because the cost of doing things wrong exceeds the cost of prevention, and promoted zero defects. Armand Feigenbaum coined total quality control. Kaoru Ishikawa gave the cause-and-effect diagram, the seven basic tools and quality circles. Genichi Taguchi defined quality loss as a function of deviation from target, usually modelled as a quadratic loss L = k(y − T)², so that even in-specification product carries a cost when it is off target.
The cost of quality is usually split, following the prevention-appraisal-failure model, into prevention costs (training, supplier development, preventive maintenance, process design), appraisal costs (testing, inspection, calibration, audits), internal failure costs (reprocessing, downgrading, scrap and lost milk solids found before dispatch) and external failure costs (complaints, returns, recalls, lost customers). Spending more on prevention usually lowers the total. In dairying, external failure through a recall of contaminated product can cost far more than years of prevention.
ISO 9001 specifies requirements for a quality management system (QMS) that an organisation can be certified against. The 2015 edition follows the harmonised high-level structure (often called Annex SL) shared by ISO management-system standards, with ten clauses: 1 scope, 2 normative references, 3 terms and definitions, 4 context of the organisation, 5 leadership, 6 planning, 7 support, 8 operation, 9 performance evaluation and 10 improvement. Clauses 4 to 10 contain the requirements. The structure maps onto Plan-Do-Check-Act: clauses 4 to 7 plan, 8 does, 9 checks and 10 acts.
Key ideas of the 2015 edition are the process approach (managing activities as interrelated processes with inputs, outputs, owners and measures), risk-based thinking (identifying risks and opportunities and acting on them, which replaced the separate "preventive action" clause of earlier editions), stronger leadership accountability, and "documented information", a single term that replaced the older split between documents and records. ISO 9000 sets out seven quality management principles on which the standard rests: customer focus, leadership, engagement of people, process approach, improvement, evidence-based decision making and relationship management. In 2024 an amendment added a requirement to consider whether climate change is a relevant issue when determining the organisation's context; the core requirements were not otherwise changed. Check the current edition before citing clause wording, because the standard is periodically revised.
ISO 9001 is not a food-safety standard. A dairy usually combines it with a food safety management system such as ISO 22000 or a GFSI-benchmarked scheme (see dairy-food-safety-systems). Because the standards share the high-level structure, one integrated manual can cover context, leadership, documented information, internal audit and management review once, with food-safety-specific clauses added. Typical dairy processes mapped under ISO 9001 are milk procurement, reception and testing, processing, packing, cold storage and dispatch, laboratory testing, maintenance, purchasing of packaging and ingredients, and customer complaint handling.
Certification is carried out by a certification body, which should itself be accredited by a national accreditation body; ISO writes standards but does not certify anyone. A certification cycle normally runs three years, with an initial two-stage audit and surveillance audits in between. Useful quality objectives for a dairy are measurable and owned: reception rejection rate, first-pass yield at the filler, complaints per million packs, laboratory turnaround time, percentage of on-time dispatch and milk solids loss to effluent.
Total quality management (TQM) is a management philosophy rather than a certifiable standard. It makes quality the responsibility of everyone, from the board to the reception platform, and seeks continual improvement of every process through customer focus, fact-based decisions and employee involvement. Its typical elements are top-management commitment, customer focus (internal customers included: the packing section is the customer of the processing section), continuous improvement (kaizen, small steady gains), employee empowerment and quality circles (small groups of workers who meet regularly to solve problems in their own area, an idea Ishikawa promoted in Japan), supplier partnership, and measurement.
Deming's chain reaction summarises the economic argument: improving quality reduces rework, mistakes and delays, which improves productivity, lowers cost, captures the market and keeps the business in business. In a cooperative dairy the supplier partnership element reaches back to the farms: training producers in clean milk production, paying for quality as well as quantity, and supporting bulk milk coolers are TQM applied to the milk shed. The general theory of TQM, BIS, AGMARK and ISO certification is covered in agribusiness quality-management-standards; this subject applies it to milk.
Ishikawa grouped seven simple tools that, he argued, could solve most quality problems on the shop floor: the check sheet, the Pareto chart, the cause-and-effect diagram, the histogram, the scatter diagram, the control chart, and stratification (some lists put the flow chart or a run chart in place of stratification). They need little training and work best when used by the people who run the process.
A check sheet is a structured form for recording data where they arise: a filler operator ticks each type of pouch defect against the hour, so that patterns by time and type are visible without later sorting. A histogram shows the distribution of a measured variable, such as fill weights or the fat percentage of standardised milk, revealing its centre, spread and shape and whether it sits inside the specification. A scatter diagram plots one variable against another, such as the temperature of milk at reception against its bacterial count, to suggest whether they are related; correlation is not proof of cause. Stratification splits data by source, for example by collection route, shift, silo or filling head, so that differences hidden in the pooled data become visible: a high average somatic cell count may come from a handful of villages.
The Pareto chart is a bar chart of problem categories sorted from largest to smallest, with a cumulative-percentage line. Juran named it after the economist Vilfredo Pareto and used it to separate the vital few causes from the trivial many; the rough rule is that about 80% of effects come from about 20% of causes, though the split is an observation, not a law. The method is: choose a measure (count, cost or milk lost), collect data for a fixed period, sort the categories in descending order, compute each category's share and the running cumulative share, and attack the categories that make up most of the total.
For example, a pouch line records 250 defects in a week: 120 seal leaks, 60 short fills, 40 print smears, 20 dents and 10 others. Seal leaks are 48% and seal leaks plus short fills are 72%, so the first improvement project should target sealing jaws and film. Pareto analysis by cost can give a different order from analysis by count: a single recall-triggering defect may outweigh hundreds of cosmetic ones, so the measure must match the decision.
The cause-and-effect diagram, also called the Ishikawa or fishbone diagram, places the problem (the effect) at the head and arranges possible causes on branches. The usual manufacturing headings are the six Ms: man (people), machine, method, material, measurement and mother nature (environment). For "high total plate count in pasteurised milk" the branches might hold untrained CIP operators; a failing flow diversion valve or worn plate gaskets; an inadequate holding time or a wrong cleaning sequence; poor raw milk; an uncalibrated thermometer or a contaminated sampling tap; and high ambient temperature in the filling room. The diagram organises brainstorming; it does not prove anything until causes are tested with data.
A flow chart maps the steps of a process with standard symbols: an oval for start and end, a rectangle for an activity, a diamond for a decision and arrows for flow. Mapping milk from reception through storage, standardisation, pasteurisation, filling and cold storage shows where samples are taken, where hand-offs happen and where contamination can enter; the same map is a mandatory preliminary step of HACCP. The "5 Whys" technique asks why a problem occurred, then why that cause occurred, usually about five times, to move from symptom to root cause. Ishikawa's group also describe "seven new management tools" for planning, such as the affinity diagram, tree diagram and matrix diagram; they are used less on the shop floor.
Every process varies. Shewhart distinguished common (chance) causes, the many small sources of variation built into the system, from special (assignable) causes, identifiable events such as a worn filling valve, a change of milk source or a wrong setting. A process with only common-cause variation is in statistical control: its output is predictable within limits. A control chart plots a statistic from successive samples against a centre line and control limits, usually set at three standard errors either side of the mean. A point outside the limits, or a non-random pattern inside them, signals a special cause to be found and removed. Control limits describe what the process does; specification limits describe what the customer wants. The two must never be confused, and specification limits are not drawn on an X-bar chart.
With three-sigma limits and a normal, in-control process, about 0.27% of points fall outside the limits by chance, so the average run length between false alarms is about 1/0.0027 ≈ 370 samples. Samples are drawn as rational subgroups: units made close together in time, so that variation within the subgroup reflects common causes and shifts appear between subgroups. Chart set-up has two phases. In Phase I historical data are charted, special causes investigated and removed, and limits fixed; in Phase II the fixed limits are used to monitor production. The theory is developed fully in agri-statistics statistical-quality-control.
For measured characteristics the mean and the spread are charted together. With subgroups of size n, the X-bar and R chart uses the grand mean X̿ and the average range R̄: X-bar chart limits are X̿ ± A2·R̄; the R chart has upper limit D4·R̄ and lower limit D3·R̄. The process standard deviation can be estimated as R̄/d2. For n = 5 the standard constants are A2 = 0.577, D3 = 0, D4 = 2.114 and d2 = 2.326; for n = 4 they are A2 = 0.729, D3 = 0, D4 = 2.282 and d2 = 2.059. The range is efficient for small subgroups; for larger subgroups (commonly above about ten) the X-bar and s chart is preferred, with limits X̿ ± A3·s̄ for the mean and B3·s̄ to B4·s̄ for the standard deviation (for n = 5, A3 = 1.427, B3 = 0, B4 = 2.089).
When only one value is available per period, for example one fat test per batch of standardised milk or one moisture result per churning of butter, an individuals and moving-range (I-MR) chart is used. The moving range is the absolute difference between successive values; σ is estimated as MR̄/1.128, so the individuals limits are X̄ ± 2.66·MR̄ and the moving-range chart has an upper limit of 3.267·MR̄. Individuals charts are sensitive to non-normal data. Counts such as somatic cells or bacteria are strongly right-skewed and are usually log-transformed before charting. Typical dairy variables charted are pouch or cup fill weight, fat and SNF of standardised milk, butter moisture and salt, powder moisture and bulk density, and pasteuriser holding temperature.
Attribute charts count defectives or defects. The p chart plots the fraction defective in subgroups of size n with limits p̄ ± 3√(p̄(1 − p̄)/n); the np chart plots the number defective with limits np̄ ± 3√(np̄(1 − p̄)). The c chart plots the number of defects per inspection unit of constant size, with limits c̄ ± 3√c̄, based on the Poisson distribution; the u chart plots defects per unit when the inspected quantity varies, with limits ū ± 3√(ū/n). A negative lower limit is set to zero. In a dairy, the p chart suits leaking or underweight pouches per hourly sample, the c chart suits print or seal faults per case, and the u chart suits defects per pallet when the number of cases varies.
Points inside the limits can still show non-randomness. The Western Electric rules flag a special cause when one point falls beyond three sigma; two of three successive points fall beyond two sigma on the same side; four of five successive points fall beyond one sigma on the same side; or eight successive points fall on the same side of the centre line. Nelson's set adds rules such as six points in a row steadily increasing or decreasing and fourteen points alternating up and down, and uses nine points on one side. Each added rule raises sensitivity but also the false-alarm rate, so a plant should state which rule set it uses. For small sustained shifts, such as a slow drift in a fat analyser, cumulative sum (CUSUM) and exponentially weighted moving average (EWMA) charts detect change sooner than a Shewhart chart.
A process in statistical control can be compared with its specification. The potential capability index Cp = (USL − LSL)/6σ compares the width of the specification with the natural spread of the process, ignoring where the process is centred. The actual capability index Cpk = min[(USL − μ)/3σ, (μ − LSL)/3σ] measures the distance from the mean to the nearer limit in units of three sigma, so it falls when the process drifts off centre. Cp = Cpk only when the process is exactly centred. Cpm (the Taguchi index) = (USL − LSL)/(6√(σ² + (μ − T)²)) penalises deviation from a target T, in line with Taguchi's loss idea. For a one-sided specification, such as a minimum fat content or a maximum moisture, only the relevant half of Cpk is used.
Capability indices assume the process is stable and roughly normal; an index computed from an out-of-control process predicts nothing. Cp and Cpk use the within-subgroup estimate of σ (for example R̄/d2), describing what the process could do; Pp and Ppk use the overall standard deviation of all data and describe what it actually did over the period. A Cpk of 1.00 means the nearest specification limit is three sigma from the mean, giving about 0.135% out of specification on that side. Customers commonly ask for Cpk of at least 1.33 (four sigma to the nearer limit) for established processes and higher for critical characteristics; the required value is a contract matter, not a universal rule.
Six Sigma practice expresses performance as defects per million opportunities (DPMO): DPMO = defects ÷ (units × opportunities per unit) × 1,000,000. The opportunities must be defined before counting; a pouch might have three: seal, fill weight and print. DPMO is converted to a "sigma level" using the normal distribution. By convention, Six Sigma tables include a 1.5-sigma long-term shift of the mean, so a six-sigma process corresponds to about 3.4 DPMO, five sigma to about 233, four sigma to about 6,210 and three sigma to about 66,807. Without the shift, a centred six-sigma process would produce only about 0.002 defects per million on both sides together. Whether the shift is applied must always be stated, because the same DPMO gives a different sigma level under the two conventions. The shift is an empirical allowance, not a law of nature, and has been criticised.
Fill weight is the clearest dairy example. Packaged-commodity laws in most countries require that the average quantity is not below the declared quantity and limit the share of packs below a tolerable negative error; the limits differ by country and should be read from the current national legal metrology rules. If the filler's σ cannot be reduced, the plant must raise the target mean to keep short packs rare, giving away product on every pack. Reducing σ, by maintaining filling valves, stabilising product temperature and level in the balance tank, is cheaper in the long run than give-away. A capability study shows how much give-away is forced by the current spread: to reach Cpk = 1.33 against a lower limit, the mean must sit at least 4σ above it.
Fat and SNF standardisation is the second example. Legal minimums for each milk type and the plant's payment-sensitive targets create tight specifications; in-line standardisation with density and fat sensors gives a much smaller σ than batch standardisation and so a higher Cpk at a lower average fat give-away. Butter moisture (bounded by a legal maximum and by yield, since each tenth of a percent of moisture is product sold), powder moisture and the holding temperature of a pasteuriser are other characteristics where a capability study tells management whether the process, not the operator, is the limitation.
Acceptance sampling decides whether to accept or reject a lot, such as a consignment of packaging film, a day's production of milk powder or a delivery of stabiliser, by inspecting a sample. It is used when testing is destructive, expensive or slow, or when 100% inspection is impractical. It controls the lot decision; it does not control or improve the process, and it cannot "inspect quality in". The acceptable quality level (AQL) is the worst process average quality, as percent defective, that is considered satisfactory for sampling purposes; a plan should accept lots at the AQL most of the time. The producer's risk (α) is the probability of rejecting a lot whose quality is at the AQL; a common design value is 5%. The limiting quality (LQ), also called the lot tolerance percent defective (LTPD) or rejectable quality level, is a poor quality the consumer wants rejected; the consumer's risk (β) is the probability of accepting a lot at that quality, commonly designed at 10%.
A single sampling plan by attributes is defined by the sample size n and the acceptance number c: inspect n units and accept the lot if the number of defectives is c or fewer. Double, multiple and sequential plans take a first small sample and decide, or take further samples when the result is borderline; they reduce the average amount of inspection at the cost of more complex administration.
The operating characteristic (OC) curve plots the probability of accepting a lot (Pa) against the lot's true fraction defective p. For a large lot, Pa is the binomial probability of finding c or fewer defectives in n units; when n is large and p small, the Poisson approximation with λ = np is convenient: for c = 0, Pa = e^(−np); for c = 1, Pa = e^(−np)(1 + np). An ideal OC curve would accept every lot better than the AQL and reject every lot worse, which only 100% error-free inspection achieves. Increasing n while keeping c/n constant makes the curve steeper and more discriminating; increasing c at fixed n moves the curve to the right and makes the plan more lenient.
When rejected lots are 100% inspected and defectives replaced (rectifying inspection), the average outgoing quality (AOQ) is approximately Pa × p for large lots; its maximum over all p is the average outgoing quality limit (AOQL), the worst long-run average quality leaving inspection. The average total inspection (ATI) counts units inspected per lot including the screening of rejected lots. These measures matter when a dairy screens rejected packaging lots itself rather than returning them.
ISO 2859-1 gives sampling schemes for inspection by attributes indexed by AQL, derived from the US military standard MIL-STD-105 and closely matching ANSI/ASQ Z1.4. The sample size code letter depends on lot size and inspection level (general levels I, II and III, with II normal; special levels S-1 to S-4 for costly tests). Switching rules move between normal, tightened and reduced inspection: under normal inspection, two lots not accepted out of five or fewer consecutive lots move inspection to tightened, and five consecutive lots accepted under tightened inspection return it to normal. ISO 3951 gives schemes for inspection by variables (from MIL-STD-414); they need smaller samples for the same protection but assume a normal distribution and apply to one characteristic at a time. Codex guideline CAC/GL 50 (General Guidelines on Sampling) explains sampling for food standards. The tables themselves should be read from the current standards.
Microbiological sampling plans of the type set out by the International Commission on Microbiological Specifications for Foods (ICMSF) are acceptance plans for counts. A two-class plan uses n, c and a limit m; a three-class plan adds a higher limit M, accepting a lot if no result exceeds M and no more than c results lie between m and M. Their legal use is covered in dairy-standards-law-worldwide. The statistics carry a hard lesson: when a pathogen is present in only a small fraction of units, a plan of ten samples with c = 0 accepts most contaminated lots (at 1% prevalence, Pa = 0.99¹⁰ ≈ 0.90). Product safety must come from process control and HACCP, with lot testing as verification.
Six Sigma, developed in electronics manufacturing in the 1980s, is a structured, data-driven approach to reducing variation and defects. Projects follow DMAIC: Define the problem, customer and critical-to-quality (CTQ) characteristics; Measure current performance and check the measurement system; Analyse data to find root causes; Improve by testing and implementing solutions; Control by standardising the new method and monitoring it with control charts. New products or processes use DMADV (define, measure, analyse, design, verify), also called design for Six Sigma. Practitioners are graded by training as Yellow, Green and Black Belts, with Master Black Belts coaching and a champion from management owning the project.
Dairy projects suited to DMAIC include reducing the standard deviation of fat in standardised milk to cut give-away, reducing pouch leakers, reducing butter moisture variation, shortening laboratory turnaround time and cutting milk solids lost to effluent. A measurement system analysis (gauge repeatability and reproducibility) is an essential early step: if the fat analyser or checkweigher varies as much as the product, no process improvement can be shown.
Lean production grew out of the Toyota Production System. It defines value from the customer's view and removes waste (muda). The seven classic wastes are transport, inventory, motion, waiting, overproduction, over-processing and defects (remembered as TIMWOOD); many add an eighth, unused human talent. Core methods are value stream mapping (drawing material and information flow from supplier to customer with times at each step), 5S workplace organisation (sort, set in order, shine, standardise, sustain), single-minute exchange of dies (SMED) to shorten changeovers, poka-yoke (error-proofing devices that make a mistake impossible or obvious), kanban pull signals, standard work and kaizen events.
Value stream maps show how little of the lead time adds value. Process cycle efficiency = value-added time ÷ total lead time; for a cultured product that spends three hours being processed and fermented within a sixty-hour lead time from milk reception to dispatch, it is 5%. Waiting in silos, cold stores and quarantine for laboratory release is often the largest share. In a dairy, product changeovers on fillers and the cleaning between products are the main SMED targets: separating internal from external tasks, pre-staging packaging, pigging lines to recover product and sequencing products from plain to flavoured or low-fat to high-fat cut both downtime and solids loss.
Overall equipment effectiveness (OEE) measures how well a machine's planned production time is used: OEE = availability × performance × quality. Availability is run time ÷ planned production time (losses from breakdowns, changeovers and CIP within the planned time); performance is actual output ÷ (run time × ideal rate) (losses from minor stops and slow running); quality is good output ÷ total output (losses from rejects and start-up waste). A filler available 90% of planned time, running at 85% of its rated speed and producing 98% good packs has OEE = 0.90 × 0.85 × 0.98 ≈ 75%. A figure of 85% is often quoted as world class for discrete manufacturing, but benchmarks vary by industry and by how planned time is defined; in dairies, cleaning schedules make comparisons between plants hard unless the definition is agreed.
Lean Six Sigma combines lean's focus on flow and waste with Six Sigma's statistical control of variation. Both depend on sustained management support and on control plans that keep gains. Poka-yoke examples in dairies include fixed-volume dosing of cultures, colour-coded hoses and fittings that prevent cleaning solution being routed to product lines, interlocked flow diversion valves and checkweighers that reject underweight packs automatically. General operations theory (capacity, scheduling, inventory, maintenance) is in agribusiness production-operations-management.
A dairy usually runs a reception (platform) laboratory beside milk intake for rapid tests, and a main laboratory for chemistry and microbiology. The microbiology area should be physically separate from the chemistry area and from production, with controlled access, surfaces that can be disinfected, hand-wash basins, and a flow that keeps clean media preparation apart from incubated and contaminated material. Pathogen work needs biosafety containment matched to the organism. Brucella from raw milk is a well-known cause of laboratory-acquired infection, and work with its cultures requires a higher containment level than routine indicator-organism counts. Many plant laboratories therefore send pathogen confirmation to a specialist laboratory. Contaminated material is decontaminated, normally by autoclaving, before disposal.
Instruments need stable power, temperature and humidity; balances need vibration-free benches away from draughts. Laboratory water is graded by ISO 3696 into grades 1, 2 and 3, grade 1 being the purest, used for sensitive instrumental methods such as liquid chromatography, and grade 3 for general washing and many wet-chemical tests. Every measuring instrument whose result matters (balances, pipettes, thermometers, butyrometers, pH meters, incubators, spectrometers) needs a calibration and verification schedule traceable to national standards, with records.
Under the Globally Harmonized System of Classification and Labelling of Chemicals (GHS), suppliers classify hazards and communicate them through labels and safety data sheets. A GHS label carries pictograms, a signal word ("Danger" for more severe hazards, "Warning" for less severe), hazard statements (H-codes) and precautionary statements (P-codes). A safety data sheet (SDS) has sixteen standard sections, from identification and hazards through first aid, fire-fighting, handling and storage, exposure controls and PPE, to disposal, transport and regulatory information. Every chemical in the laboratory should be on an inventory with its SDS accessible to staff; adoption of GHS and the exact form of national rules vary by country.
Dairy methods use several hazardous chemicals. Concentrated sulphuric acid in the Gerber fat test is corrosive and reacts violently with water, so acid is added to water, never water to acid, and butyrometers are handled with face and hand protection; the centrifuge must be balanced and its lid closed while running because butyrometers can break. Isoamyl alcohol is flammable and harmful. The Röse-Gottlieb reference method uses ammonia, ethanol, diethyl ether and light petroleum: the ethers are extremely flammable, their vapours are heavier than air, and diethyl ether forms explosive peroxides on storage, so it is dated on opening, tested or discarded on schedule and handled in a fume cupboard away from ignition sources. Kjeldahl digestion releases acid fumes and needs extraction. Some preservatives used for composite milk samples, such as potassium dichromate, are toxic and carcinogenic, which is why many laboratories have moved to less hazardous preservatives; check what the method and national rules permit.
Chemicals are stored by compatibility, not alphabetically: acids apart from bases, oxidisers apart from flammables and organic materials, and flammable solvents in approved flammable-storage cabinets in quantities kept to the working minimum. Concentrated acids are stored low, in secondary containment. Fume cupboards are tested for face velocity on a schedule. Eye-wash stations and safety showers are within easy reach of acid and alkali work and are flushed regularly; spill kits suited to acids, alkalis and solvents are stocked and staff are trained to use them.
Waste is segregated at source: halogenated and non-halogenated solvents, acids and alkalis for neutralisation, heavy-metal wastes and biological waste are collected separately and disposed of under national rules. Personal protective equipment (laboratory coats, safety glasses or face shields, suitable gloves) is the last line of defence after elimination, substitution, engineering controls and administrative controls in the hierarchy of controls. A laboratory safety plan names responsibilities, induction training, emergency contacts and incident reporting; good safety records also support accreditation, since ISO/IEC 17025 requires suitable facilities and environmental conditions.
ISO/IEC 17025 specifies the general requirements for the competence, impartiality and consistent operation of testing and calibration laboratories. The 2017 edition has five requirement clauses after the introductory ones. Clause 4, general requirements, covers impartiality and confidentiality. Clause 5 covers structural requirements: legal identity, defined scope and management responsibility. Clause 6 covers resources: personnel competence, facilities and environmental conditions, equipment, metrological traceability, and externally provided products and services. Clause 7 covers processes: 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 (including statements of conformity using a stated decision rule), complaints, nonconforming work, and control of data and information management. Clause 8 covers the management system, which can be met by option A (the clauses of the standard itself) or option B (a management system operated to ISO 9001 that also supports the laboratory requirements).
Like ISO 9001:2015, the 2017 edition adopted risk-based thinking and became less prescriptive about documents. Unlike ISO 9001 certification, accreditation to ISO/IEC 17025 assesses technical competence for each specific test listed in the laboratory's scope, so a laboratory may be accredited for fat by the Röse-Gottlieb method but not for aflatoxin M1.
Accreditation is granted by a national accreditation body that itself conforms to ISO/IEC 17011. Examples are NABL in India, UKAS in the United Kingdom, DAkkS in Germany, IANZ in New Zealand, NATA in Australia, CNAS in China, and several bodies in the United States, including A2LA and ANAB. Bodies that pass peer evaluation sign the International Laboratory Accreditation Cooperation (ILAC) Mutual Recognition Arrangement, under which test reports from an accredited laboratory in one signatory economy are accepted as equivalent in others. This matters for dairy exports, where an importing authority may require results from an accredited laboratory or from one it officially designates.
Accreditation should not be confused with certification: a certification body certifies that a management system meets ISO 9001 or ISO 22000; an accreditation body attests that a laboratory, or a certification body, is competent for defined tasks. Regulators often add their own approval. Food laboratories testing official samples may need notification or recognition by the food authority, and in the United States laboratories testing Grade A milk under the cooperative interstate programme are evaluated and certified under the FDA and NCIMS laboratory programme; see dairy-standards-law-worldwide.
Laboratories prefer standard methods published by ISO and IDF jointly, AOAC International, national standards bodies or the food regulator's manual. A standard method used as written must be verified: the laboratory shows it can achieve the published performance in its own hands. A non-standard, modified or in-house method must be validated by establishing its performance characteristics: selectivity, limit of detection and limit of quantification, linearity and working range, trueness (bias, often by recovery of a spike or by analysis of a certified reference material), precision (repeatability within a run, intermediate precision across days and analysts, reproducibility across laboratories), robustness, and measurement uncertainty. A common convention estimates the limit of detection as about three times, and the limit of quantification as about ten times, the standard deviation of blank results; conventions differ, so the one used must be stated.
Measurement uncertainty is estimated by the approach of the Guide to the Expression of Uncertainty in Measurement (GUM) or from validation and proficiency data. Each component is expressed as a standard uncertainty; Type A components are evaluated statistically from repeated measurements, Type B by other means such as calibration certificates or tolerances. Independent components combine as the root of the sum of squares to give the combined standard uncertainty uc; the expanded uncertainty U = k·uc, with a coverage factor k = 2 giving roughly 95% coverage for a normal distribution. From collaborative-study precision data, the repeatability limit r = 2.8·sr and the reproducibility limit R = 2.8·sR give the largest difference expected, at 95% probability, between two results obtained under those conditions; 2.8 is 1.96 × √2.
Both ISO 9001 and ISO/IEC 17025 require planned internal audits and periodic management review. ISO 19011 gives guidance on auditing management systems: audit programmes, auditor competence, planning, collecting objective evidence by interview, observation and record review, and reporting findings. In a laboratory three audit styles complement each other. A horizontal audit examines one requirement across many tests (for example equipment calibration for all instruments). A vertical audit follows one sample from receipt through testing and calculation to the issued report, checking every record on the way. A witness audit watches an analyst perform a test to see whether practice matches the written method.
Findings are graded. A major nonconformity is the absence or total breakdown of a required element, or a failure that casts doubt on the validity of results; a minor nonconformity is an isolated lapse that does not undermine the system; an observation or opportunity for improvement notes a risk without a breach. Each nonconformity needs correction (fixing the immediate problem), root-cause analysis and corrective action to stop recurrence, with a check that the action worked. If a nonconformity may have affected reported results, the laboratory must evaluate the impact and inform customers where necessary. Management review considers audit results, proficiency performance, complaints, workload, resources and improvement actions.
In proficiency testing (PT), a provider sends the same homogeneous, stable material to many laboratories and compares their results with an assigned value. PT providers are accredited to ISO/IEC 17043, and the statistical methods are set out in ISO 13528. The most common performance score is the z-score: z = (x − X)/σpt, where x is the laboratory's result, X the assigned value and σpt the standard deviation for proficiency assessment, set by the provider. The usual interpretation is |z| ≤ 2.0 satisfactory, 2.0 < |z| < 3.0 questionable (a warning signal) and |z| ≥ 3.0 unsatisfactory (an action signal). A laboratory that reports its own uncertainty can also be scored with the En number, En = (x − X)/√(U_lab² + U_ref²) using expanded uncertainties, where |En| ≤ 1 is satisfactory; zeta scores use standard uncertainties in a similar way.
A single questionable score calls for review; two successive questionable scores in the same direction, or one unsatisfactory score, call for investigation and corrective action. Accreditation bodies expect laboratories to take part in PT for each accredited test where a scheme exists. Dairy PT schemes run for fat, protein, lactose, total solids, freezing point, somatic cell count, bacterial counts, inhibitors and residues. Where no scheme exists, interlaboratory comparisons with another competent laboratory or analysis of certified reference materials (produced under ISO 17034) give external evidence.
Between external checks, the laboratory monitors itself. Control samples or reference milks are analysed with each batch and plotted on Shewhart charts; results outside the action limits stop reporting until the cause is found, and drift within the limits prompts recalibration. Duplicate analyses check repeatability against the method's repeatability limit; blanks check contamination; spiked samples check recovery in residue methods.
Milk payment laboratories rely on mid-infrared analysers for fat, protein, lactose and solids and on flow cytometers for somatic cells and bacteria. These routine instruments are indirect methods and must be calibrated against reference methods: Röse-Gottlieb (or Gerber where the law allows it for routine use) for fat, Kjeldahl for protein, and the direct microscopic method for somatic cells. ISO 9622 (IDF 141) gives guidance on the use of mid-infrared spectrometry for milk, and ISO 8196 (IDF 128) covers the definition and evaluation of the overall accuracy of alternative methods. Calibration sets of milks with reference values, often supplied by a central laboratory, and regular slope and intercept checks keep many instruments in a region aligned, so that farmers are paid fairly wherever their milk is tested.
Shelf life is the period during which a product, stored under stated conditions, stays safe and keeps its intended sensory, chemical, physical and microbiological characteristics and any labelled nutritional claims. The first step in any study is to identify the limiting mode of deterioration, because each product fails in its own way. Pasteurised milk is usually limited by bacterial spoilage: Gram-negative psychrotrophs introduced after pasteurisation (post-process contamination) cause rapid spoilage, while in well-sealed systems heat-resistant, cold-tolerant spore-formers such as Bacillus species set the end point. Extended shelf life (ESL) milk relies on higher heat treatment or microfiltration and very clean filling. UHT milk is microbially stable when sterile but is limited by physical and chemical change: age gelation, sedimentation, fat separation, Maillard browning and stale or oxidised flavours, often accelerated by heat-stable proteases and lipases of psychrotrophs that grew in the raw milk. Yoghurt is limited by post-acidification, syneresis and yeasts and moulds; soft cheese by moulds and, for safety, Listeria monocytogenes; butter by oxidation, lipolysis and moulds; milk powder by lipid oxidation, Maillard browning and caking linked to moisture and glass transition.
Codex labelling texts distinguish a "use-by" date, after which a perishable product should not be sold or eaten for safety reasons, from a "best before" date, the end of the period of optimum quality. National rules on date marking differ; see dairy-standards-law-worldwide.
A real-time storage study keeps samples under the intended conditions, often with a realistic temperature-abuse scenario for chilled products, and tests them at intervals for the limiting attributes: counts, pH, titratable acidity, peroxide value, sensory scores and appearance. The end-point must be defined in advance: a count limit, a sensory score below which a panel rejects the product, or a chemical threshold. Sensory shelf life is increasingly estimated by survival analysis of consumer rejection, which asks at what storage time a stated proportion of consumers would reject the product. Studies need several production batches, because batch-to-batch variation in raw milk and hygiene is often larger than the trend.
For ready-to-eat foods that can support Listeria monocytogenes, such as soft cheeses, food-safety authorities expect the operator to show that the pathogen will not exceed the legal limit before the end of shelf life. Durability studies sample naturally contaminated product over storage; challenge tests inoculate product with the organism and measure its growth; predictive microbiology models estimate growth from pH, water activity, temperature and preservatives. Technical guidance on how to design such studies is published by regulators and reference laboratories and should be followed in its current version.
Accelerated shelf-life testing (ASLT) stores product at higher temperature (or humidity) to speed the limiting reaction, then extrapolates to normal storage using a kinetic model. Quality loss often follows zero-order kinetics (constant rate, linear change, typical of browning or sensory score) or first-order kinetics (rate proportional to the remaining quantity, typical of vitamin loss). For zero order, time to end-point = (initial value − limit)/rate; for first order, t = ln(C0/C)/k, so the time to lose half is ln2/k. The temperature dependence is described by the Arrhenius equation k = A·exp(−Ea/RT), or more simply by Q10, the factor by which the rate rises for a 10 °C increase. With a known Q10, shelf life at a lower temperature = shelf life at the test temperature × Q10^(ΔT/10). A Q10 of 2 to 3 is common for many chemical reactions, but the true value must be measured for the product, because a small error in Q10 becomes a large error after extrapolation over 20 or 30 degrees.
ASLT works best for shelf-stable products whose limiting reaction stays the same over the temperature range: lipid oxidation in whole milk powder, browning in UHT milk or powders, oxidation in butter oil or ghee. It is unreliable when higher temperature changes the mechanism: fat melting in butter or chocolate-coated ice cream, crossing the glass transition in powders, or a change in which micro-organisms grow in chilled products. For chilled pasteurised milk and fresh cheeses, real-time studies at the intended temperature, supported by predictive models, remain the norm. Shelf-life claims are verified after launch by testing retained samples and by monitoring complaints. General theory is covered in food-technology's shelf-life-stability and quality-assurance-lab-accreditation subjects.
Learn the definitions that recur in every paper (QC versus QA, AQL, producer's and consumer's risk, Cp versus Cpk, accreditation versus certification) and practise the hand calculations until they are quick: X-bar–R and p-chart limits, Cp and Cpk, Poisson acceptance probabilities, z-scores, combined uncertainty and Q10 extrapolation. In in-plant training, keep a control chart of real fill weights or fat tests for a week and compare it with the specification; it teaches more than any table. Sit with the laboratory's quality manager during an internal audit.
A QA officer or laboratory analyst five years in runs the daily system: charts fill weights and standardisation results, reacts to out-of-control signals, maintains calibration and internal QC, prepares for NABL or other accreditation surveillance, investigates proficiency results outside |z| = 2, and drafts shelf-life studies for new products. Typical decisions: whether a signal is a special cause or a measurement problem, whether a lot can be released, and which corrective action removes a root cause rather than a symptom. Sound statistics turns arguments with production into evidence.
A head of quality with 25 years' experience sets the quality policy and integrates ISO 9001, food-safety certification and laboratory accreditation into one system that auditors and customers trust. Decisions are about money and risk: how much to invest in in-line analysers to cut give-away, whether to outsource pathogen testing, which customers need Cpk evidence, how to structure a recall-proof traceability and release system, and how to build a quality culture across procurement, plants and distribution. They present cost-of-quality data to the board.
A senior scientist or professor works on the methods behind the system: validating rapid microbiological and spectroscopic methods against reference methods, designing and analysing proficiency schemes, modelling shelf life with kinetic and predictive-microbiology tools, and serving on IDF, ISO or national standards committees that write sampling plans and methods. They teach SQC and quality management at postgraduate level, set examination papers, and review whether accepted conventions (the 1.5-sigma shift, Q10 extrapolation, small-n sampling for pathogens) are being used beyond their evidence.
Sampling-plan tables of ISO 2859-1 and ISO 3951 and control-chart constants beyond those quoted are not reproduced; read them from the current standards or a statistics text. Legal limits (fill-weight tolerances, microbiological criteria, residue limits, date-marking rules) are deliberately not quoted because they differ by country and change; they belong to dairy-standards-law-worldwide and current regulations. Laboratory biosafety levels, exposure limits and storage quantities for chemicals are governed by national rules not set out here. Measurement-system analysis (gauge R&R) and design of experiments are named but not worked through. Multivariate SPC, process analytical technology with in-line spectroscopy, and statistical treatment of censored microbiological data are not yet covered.
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. Legal limits are deliberately not tested; read them from current regulations.
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.