What the module does

The module collects characteristic data at the right frequency, charts it in real time with configurable run rules, forces a documented reaction to every signal, computes capability correctly, and manages inspection plans, sampling, gauge studies and first article inspection. It replaces the real-time SPC and inspection tools a business usually pays for. The module specification names Advantive (InfinityQS ProFicient and Enact) ^[1][2], Minitab Real-Time SPC, WinSPC, SQCpack, 1factory, High QA and DataLyzer.

What those tools sell is not the picture of a chart. It is the discipline behind it: a number is entered once, tied to a lot and an order, judged by rules on every point, and when a rule fires somebody must answer. This module builds that discipline on the platform's shared spine, so parts, machines, people and lots are the same rows every other module uses.

What is in, and what is not

• In: characteristics and control plans; sampling and inspection plans for receiving, in-process, final and first article inspection; data capture by hand, gauge, machine tag or form; control charts for measured and counted data, including EWMA and CUSUM, with Western Electric and Nelson run rules; a required reaction to every signal with assignable cause and corrective action codes, and escalation to an NCR; capability (Cp, Cpk, Pp, Ppk, and non-normal methods); measurement system analysis; and AS9102 first article reports.
• Out: laboratory sample management (module M10) and the NCR and CAPA workflow (module M08). When a signal needs a nonconformance record, this module escalates and keeps a link; the record itself belongs to M08.

A characteristic is the thing you measure

A characteristic is one measured or counted feature of a part or process: a bore diameter, a seal temperature, a count of scratches. Everything else in the module hangs from it. Each one has a code, a name, a type (variable for a measured number, attribute for a count or a pass-fail), a unit, and for a measured one a nominal value with a lower and an upper specification limit (lsl and usl). A special class marks the ones that matter most: none, cc (critical), sc (significant) or key. The type of gauge, the number of decimals the gauge reads and an optional machine data tag (so the value can arrive by itself) complete the row.

Specification limits say what the customer or the design will accept. They are not control limits. Control limits come from how the process behaves, and chapter 3 keeps the two apart.

A control plan says how each one is watched

A control plan is a header (a plan number, the item, a revision, a phase of prototype, pre-launch or production, and a status of draft, approved or superseded, with an optional link to the FMEA, the risk analysis it came from) and a list of lines in the AIAG format. Each line names the process step, the characteristic, the evaluation method, the sample size, the frequency, the control method and the reaction plan: what the operator does when the measurement is not what it should be. A line with no reaction plan is a chart that nobody has to answer.

Eleven capabilities in three tiers

The capability map lists what the module does, and each capability carries a tier. MVP must exist for a small or mid-size plant to cancel the incumbent subscription with confidence. STD is parity with mainstream tools. BIC marks best-in-class differences. MVP and STD are required for the course; BIC is advanced work.

The standards, and what each one is used for

Thirteen tables on the shared spine

Every table starts with the standard columns every platform table has: id, tenant_id, created_at, created_by, updated_at, updated_by, row_version, archived_at and ext. The module has no parts list, machine list, people table or lot table of its own: it points at core_item, core_equipment (machines and gauges), core_person, core_lot, core_uom, core_data_tag, core_reason_code, core_e_signature and the rest of the spine. Its own tables share the prefix spc_.

The table below is the full definition: every column the module adds to the standard ones, with its type and whether it is required (marked req). Types are written as in the specification: text, int, num (a decimal number), bool, ts (a timestamp with time zone), date, qty (a quantity), uuid and json; an arrow means a foreign key to that table; in {…} lists the only allowed values.

Typed numbers, and five soft-link columns

Readings, specification limits and control limits are numeric columns, not text and not one JSON blob. A single blob for all readings is a pitfall this course adds to the specification's three: it makes limits, charts and capability impossible to query. The decimals column on the characteristic says how many places the gauge reads, and the server rounds to it.

Five columns are soft links pointing at four tables: plain uuid columns with a comment saying which table they will point to, and no foreign key. They are operation_ref (to the MES module's mes_operation_definition), operation_run_ref (to mes_operation_run), fmea_ref (to qms_fmea) and the two event references quality_event_ref (on spc_inspection_lot) and escalated_event_ref (on spc_violation), both to qms_quality_event. The other modules may not be installed, and a foreign key to a table that does not exist would stop the migration.

Six roles, and rules for two people

The roles are jobs, not names, and each maps to a row in the spine's core_job_role. The operator measures at the machine. The inspector does receiving, in-process, final and first article inspection. The quality engineer owns characteristics, control plans, charts, studies and imports. The quality approver (usually the quality manager) approves control limits, control plans and first article reports. The supervisor answers and closes signals in their own area. The manager reads the board and changes nothing.

Row-level security is generated from the Role and Exposure Matrix, so every right a later session needs must be in the matrix first. This module adds rules that involve two different people (a design choice of this course, not a requirement of the specification): the person who acknowledged a violation may not close it; the person who proposed new limits may not approve them; an inspector may not approve their own first article report. Nobody gets delete: a correction is a new row, a new revision or the archived_at column.

Events and screens

The screens sit on the platform's three surfaces. The operator surface (a phone screen at the machine) shows the characteristic with its limits beside a large number box and a full-width Save button, and queues readings on the device when offline. The supervisor surface lists signals to answer. The manager surface is a board readable from across the room. The quality engineer's plan and chart screens, and the approver's limits screen, are desk screens. The server, not the browser, decides whether a reading is inside its specification limits.

Two kinds of variation, and two kinds of limit

Every process varies. A control chart (invented by Walter Shewhart and standardised for charts in ISO 7870-2 ^[3]) plots a statistic over time against a centre line and limits set from the process's own behaviour, usually three standard deviations (sigma) either side ^[1]. A point inside the limits with no pattern is ordinary variation: leave the process alone. A point or pattern outside what ordinary variation would produce is a signal: something changed, and there is a cause to find.

A point can be inside specification and out of control, or the reverse. The two answers are separate, and the capture route reports both.

Choosing the chart

For an X-bar and R chart the limits come from the average range multiplied by a constant that depends on the subgroup size. Those constants are printed in the AIAG SPC Reference Manual and in ISO 7870-2. Type them from the table in your copy, never from memory, and test them: the module's definition of done is that the chart constants and rule detection reproduce a published reference dataset exactly.

Run rules: reading the pattern

Walking a point past the limit is not the only sign of a change. The Western Electric handbook (1956) and Nelson (1984) list patterns that random variation rarely makes ^[1][2]. The module evaluates them on every point, in real time, and which rules are on is a setting of each chart (the rules column).

The required reaction

When a rule fires, the module inserts a spc_violation with the rule code and the time of detection, in the same database transaction as the point, so a point and its violation exist together or not at all. A chart with no approved limits collects points and evaluates nothing.

• open: a rule has fired and nobody has answered.
• acknowledged: a person has recorded an assignable cause code (a coded reason from the quality reason codes, such as tool wear or wrong material) and the corrective action taken. The database refuses an acknowledgement missing either.
• closed: a different person has reviewed the answer and closed it. The database refuses a close with no cause code and corrective action.

The median minutes from detected_at to acknowledged_at is the signal response time, the first KPI of the module. When a signal needs a hold on a lot or a nonconformance, the supervisor escalates it: an action item is created with a link back, and, where M08 is installed, the id of its quality event is stored in escalated_event_ref.

Control limits are governed

Control limits are part of the controlled process, so changing them needs approval and is audit-trailed. A limit's basis is historical (computed from a stable period), standard or manual. A proposal is made by the quality engineer, approved by a different person, the quality approver, with an e-signature, and the approve route writes the new limits together with limits_set_at, limits_set_by and limits_approval_signature_id. The database refuses any other change to those columns.

10.03 9.88 10.06 10.12 9.95 9.91 10.05 9.86 10.08 9.97 10.13 9.92
10.34 10.08 9.98 9.89 10.07 10.01 10.02 9.90 10.05 10.09 10.03 10.14
10.02 10.06 10.01 10.11 10.04 9.96 9.87 10.10 9.93 9.99 10.08 9.92

Capability: how well could the process meet the specification?

Capability compares the spread of the process with the width of the specification ^[1]. Four indices are in use. Cp and Cpk use the variation within subgroups (sigma within, estimated from the subgroup ranges or the moving range). Pp and Ppk use the overall variation of all the data (sigma overall). Cp and Pp ignore where the process is centred; Cpk and Ppk punish an off-centre mean.

Do not confuse this picture with the chart in chapter 3. Here 9.70 and 10.30 are a tighter, made-up specification chosen to make the arithmetic easy; in chapter 3 the same numbers were control limits, and the seed characteristic you load in Session 2 has a specification of 9.60 to 10.40. Specification limits come from the drawing; control limits come from the process.

• Expected ppm: the parts per million expected outside the specification, from the distribution fitted.
• Normality: the indices above assume a normal distribution. The module stores a normality test result and, when the data is not normal, the distribution or method used (the distribution column).
• Targets: the specification gives 1.33 as an example Cpk target and 1.67 for special characteristics. Falling below a target publishes spc.capability.below_target.
• Stable first: capability describes a stable process. A study is saved with stable true only if the rules engine found no signal in its period; otherwise it carries stable false and a flag on the screen.

The gauge first: measurement system analysis

Every reading carries the gauge's own variation. A measurement system analysis (MSA) measures how much, so a capability figure is not an artefact of a poor gauge. The module stores crossed and nested gage R&R studies (by ANOVA, analysis of variance, or the average-range method), bias, linearity, stability, and attribute agreement (by kappa, a measure of agreement beyond chance) ^[2]. The AIAG manual's usual layout is 10 parts, 3 operators and 3 trials, 90 readings, measured blind in random order.

%GRR is the share of the variation that belongs to the gauge and its operators. Marginal gauges need a plan (training, a better fixture or another gauge), and capability figures taken with them are marked provisional.

Inspection and sampling: deciding about a lot

Charts watch a process. Inspections decide about a lot. An spc_inspection_plan says which characteristics are checked for an item at a receiving, in-process, final, first article, audit or source inspection, and which sampling plan applies. The sampling schemes are fixed (a stated number), z14_aql and z19 (ANSI/ASQ Z1.4 for attributes and Z1.9 for variables), skip_lot, c_zero (accept only if no defect is found in the sample) and hundred_pct (every piece). The acceptance quality limit (AQL) is the quality level the Z1.4 tables treat as satisfactory. Sample sizes and accept or reject numbers come from the tables in the standard ^[5], never from memory.

First article inspection

A first article inspection (FAI) proves a new or changed part can be made to the drawing. SAE AS9102C ^[3] defines three forms: Form 1, part number accountability; Form 2, product accountability, covering raw materials, specifications, special processes and functional testing; and Form 3, characteristic accountability, verification and compatibility evaluation, in which each drawing characteristic is numbered (ballooned) and given a result. The module stores the three forms as JSON on spc_fai_report (Form 3 is required), with a status of draft, complete, approved or rejected. The inspector may not approve their own report; the quality approver signs it.

Measuring the module, and cutting over

To retire the old tool, export its characteristics, specification limits and historical measurements, and reload the history so that limits and capability can be recomputed and compared with the incumbent's own figures. That comparison is the validation check. Imported history is run through the rules engine in report-only mode, so last year's signals appear in a report and not as open violations that page the floor. Run both tools side by side for at least a week, explain every difference, and cancel only after the cut-over works and the final export is stored.