SystemManufacturing Intelligence

Process Drift Detection

Process drift detection that flags a result sliding inside its specification, finds what the drifting batches share that the steady ones do not, and leaves the watch actions to a signed assessment.

Follow one dissolution trend from signal to signed watch ↓
Watch with transcript →
The problem

Dissolution slid for months, and every batch still passed release

A result that slides inside its specification is allowed, quiet and dangerous. Each batch passes and is released, the control chart is looked at each quarter, and the annual product review may be the first place the slope is read.

When the weakest tablet finally falls under the stage-one mark, the batch goes to further testing, an out-of-specification investigation may hold it and the batches behind it, and the client's supply waits. Finding why means joining the LIMS, the batch record, the ERP and supplier certificates by hand, where two lots that arrived in the same weeks look equally guilty.

Built forProcess engineersHeads of qualityMSAT leadsSupplier quality specialists

The Problem Today

Every batch passes its specification while the results slide, so nobody looks until one fails.

12 mo
longest a trend may wait for the annual review, 211.180(e) published

Joined by hand in a spreadsheet

  • Lab results
  • Batch steps
  • Lot records
  • Suppliers
  • Inside spec, out of sightA slide within limits raises no alarm
  • Coincidence read as causeTwo lots that arrived together look alike

The Solution

  1. Drift flagged batch by batchEach approved result read against the CPV baseline and its slope
  2. What the drifting batches shareEach batch drawn to its press, crew, room, lots and supplier site
  3. A challenger corrects the draftParticle size, more than the milling site, explains the drifting batches
  4. Watch actions signed by twoThe engineer authors; the head of quality countersigns after reading

Built-in controls

  • Release stays on the specification, never on the trend
  • Steady batches read before the countersign
  • Suppliers answer in fixed fields that are checked

What You Get

Plants act on a drift while every batch still passes, with the evidence a client would ask for.

  • Caught with margin leftWatch actions in place before a stage-one failure
  • Clients told before they askWith counts both ways and the steady batches shown
  • One finding, many productsParticle size tested at receipt for each lot from that supplier
Features

Process drift detection features, from a sliding trend to signed watch actions

A process engineer, the head of quality and supplier quality follow a tablet whose dissolution at thirty minutes slid batch after batch while each one passed release.

demoThe plant, client, people, batches, lots and supplier shown are sample data.

  1. 01Drift chart

    Process drift detection on a dissolution trend that still passes

    • Batch mean and weakest tablet together. The engineer sees the mean slide and the weakest tablet close in on the stage-one mark, batch after batch.
    • Read as each result is approved. The drift watcher re-reads each watched attribute when the LIMS approves a batch's results, long before the annual review.
    • Baseline from the CPV plan. Center and alert lines come from the signed CPV plan, so the signal rests on limits QA already approved.
  2. 02Commonality lines

    What the drifting batches share that the steady ones do not

    • Counts both ways. Each factor shows how many drifting and how many steady batches it touches, so a coincidence reads as one.
    • Mixed factors stay grey. Press, crew, room humidity and blend hold split across both groups, and the screen leaves them unlit.
  3. 03Factor counts

    Batch-to-batch commonality read from the ERP lot master and certificates

    • An exact test beside each count. The engineer reads how likely the split is by chance, with the batches behind it listed.
    • Each count opens its source. The lot master and the batches behind a count open from the factor, so nothing rests on the chart alone.
  4. 04Corrected finding

    Challenger corrects the agents' finding from the mill site to particle size

    • Draft finding overturned. Two drifting batches used a coarse lot from the first site, so the finding moves from the site to particle size.
    • Confounding tested. A lubricant lot that arrived in the same weeks is checked inside its own batches and does not separate the groups.
    • Steered in the engineer's words. Leaving out the engineering batches redraws the counts, and the finding holds.
  5. 05Supplier request

    Supplier data request in fixed fields, approved by supplier quality

    • Pick lists and numbers. Milling site and screen come from fixed lists, d90 as a number and the change notice as yes or no with a reference.
    • Sent by the supplier quality specialist. The agent drafts the request, and only the supplier quality specialist approves and sends it.
  6. 06What-if

    Projection of batches left before stage-two dissolution testing

    • Each assumption editable. The slope, the tablet-to-tablet spread and the next lot's d90 can be changed, and each formula is shown.
    • Projections kept as estimates. The screen labels each projection an estimate, apart from the results it is built on.
  7. 07Trend assessment

    Trend assessment with watch actions, signed by the engineer as author

    • Each line cited. Signal, finding, projection and risk each link to the CPV plan, the LIMS, a certificate or the ERP.
    • Release stays on the specification. Dissolution profiles on the next batches are for information, so the watch never changes how a batch is released.
    • A finding that reaches other products. Particle size is tested at receipt on each lot from that supplier, with an internal alert below the limit.
  8. 08QA evidence

    Head of quality reads the steady batches before countersigning

    • The batches that stayed put. The countersigner sees the steady batches with their API lots, which is what makes the finding believable.
    • Evidence opened before signing. The steady batches, the challenger's points and the certificate are each opened before the countersignature.
  9. 09Client letter

    Client notification of a trend, checked before the head of quality sends it

    • Told before the client asks. The client hears of the trend while each batch still passes, as its quality agreement asks.
    • Unproven cause words removed. The check replaced a claim that the second site caused the drift with the association the data support.

Only the supplier quality specialist decides what the API supplier is asked

Engineering finds and authors, QA countersigns, MSAT carries the rule across products, and supplier quality asks the supplier.

Process engineers

Own a product's process performance and open trend assessments

  • A signal while batches still pass
  • What drifting batches share, counted
  • Watch actions drafted with citations
Heads of quality

Countersign trend assessments and watch actions, and decide disposition

  • The steady batches beside the drifting
  • The challenger's points answered
  • A client letter checked for unproven claims
MSAT leads

Own the watch board and the rules across plants and products

  • One set of rules for each plant
  • Signals assigned from one queue
  • A finding turned into a rule for other products
Supplier quality specialists

Own API and excipient suppliers, their agreements and data requests

  • Requests in fields a supplier can answer
  • Replies checked against certificates
  • Corrective action requests when a change went unnotified

A lubricant lot that arrived with the coarse API looked just as guilty

Particle size, more than the site

Two drifting batches used a coarse lot from the first milling site, so the draft's site finding over-reached.

A lubricant lot that only looked guilty

It arrived in the same weeks; inside its batches, the one steady batch had a fine API lot.

Over-lubrication ruled out

Blend time matched the master record on each batch, and hardness and compression force held steady.

A letter that claimed a cause

The client letter's draft blamed the second site; it now says coarser API lots, with confirmation asked of the supplier.

More screens

One watch board re-reads 612 attributes on 46 products as results are approved

Around the finding: every signal across plants, what the watch actions cost the lab, and a single batch's records.

Points from the inspector and the statisticianEach point answered by an agent with its source or sent to a person, with two catches in the agents' own drafts.
Signals across plants and clientsOpen drift, shift, watch and variability signals with what the agents saw, the owner and the assessment due date.
The lab's added workDissolution profiles and particle-size tests from the watch actions, with analyst hours, for the QC supervisor to accept or decline.
Products and attributes on one watch boardWatched attributes across products, plants and clients, with open signals, read again as each batch's results are approved.
One batch and the records behind itSix tablets at stage one, the batch mean and lowest tablet, and each factor value beside its source record.
Signatures, checks and agent steps in orderThe countersignature with its meaning, the letter check and each agent step, with its reason and session; an audit CSV and an evidence pack download it.

demoSample data, from the working solution.

The QC lab supervisor signs too, for the lab hours the watch actions add

Process engineer
Sees
The drift chart, the counts both ways and the challenger's points
Signs
The trend assessment, as its author
Head of quality
Sees
The steady batches, the challenger's points and the certificate, each opened
Signs
The countersignature: watch actions adequate to the risk
Supplier quality specialist
Sees
The fixed-field request and each field of the reply
Signs
The request before it goes, and acceptance of the answer
QC lab supervisor
Sees
The tests the watch actions add, with analyst hours
Signs
Acceptance of the extra testing into the worklist

The lab, batch and lot records a drift is read from

LIMS resultsRelease dissolution by tablet and other approved results, read as each batch is approved.
Electronic batch recordPress, punch set, crew signatures, blend times and in-process hardness.
Process historianCompression room humidity and other conditions for each batch.
ERP lot masterWhich API, lubricant and filler lots each batch used, and when they arrived.
Supplier certificatesParticle size and milling site for each API lot, as the supplier certified them.
CPV plan and proceduresThe baseline, alert limits and the assessment clock the signal is judged against.

Questions engineers and QA raise about process drift detection

What is process drift detection?

Process drift detection finds a result moving away from its baseline while it is still inside its specification. Here agents read each approved result against the CPV baseline, find what the drifting batches share, and draft a trend assessment that the process engineer signs, then the head of quality countersigns.

What is the difference between out of trend and out of specification?

Out of specification means a result failed its limit. Out of trend means results are moving in a way the baseline did not predict while each still passes, which is when a plant can still act without a failed batch.

Does a drift signal hold or reject a batch?

No. Release stays on the specification. A drift opens a trend assessment, and its watch actions, such as extra dissolution profiles, are for information unless QA decides otherwise.

How does it avoid blaming a coincidence?

Each factor shows counts for both the drifting and the steady batches with an exact test, and a challenger tests confounding, such as two lots that arrived in the same weeks, before anyone signs.

Does it replace a CPV program?

No. A CPV plan sets what is watched and its limits. This reads each result as it is approved against that plan, explains a drift from the batches' own records, and turns the answer into signed watch actions.

How early can it warn before an out-of-specification result?

As soon as the LIMS approves a batch's results, rather than at the quarterly chart review or the annual product review. A projection then estimates how many batches are left before stage-two dissolution testing, labeled as an estimate, with the slope, the tablet-to-tablet spread and the next lot's particle size open to change.

Show us the trend
that still passes.

Send thirty batches of one attribute with their lots and certificates, and we will name what the drifting ones share.