Case study, process engineering

From spreadsheet configs to a validated configuration builder

A global automotive safety system manufacturer's textile components division ran line configuration through Excel, email and manual PLM entries, with no shared source of truth, no validation and no traceability. This is the design and build of the tool that replaces it with a guided, rule-enforced workflow.

Manufacturing
Master-data validation
Azure AD and on-premise
Full version history
MVP scopeDesigned, built, not yet live
0
wizard steps, each gating the next
0
architecture zones, each reaching only the one below
0
decisions that set what the tool can be trusted to do
0
superseded options an engineer can select
The client

A textile Product Area building production lines for cushion and seam-based components across multiple manufacturing sites.

The challenge

Nothing enforced which machine, tool or poka-yoke combination was valid, so invalid configurations surfaced in review, procurement or on the shop floor.

The build

A five-step quotation wizard where each tab feeds the next and nothing generates until every mandatory field is filled and valid.

Where it stands

The MVP is designed and built. Performance testing and go-live outcomes are targets, not results, and this page says so wherever a number appears.

The client

A textiles Product Area configuring lines by hand

Central Engineering and Purchasing owned the master data. They had no system to publish it, so engineers worked from whatever copy they had.

The client's textile Product Area builds production lines for cushion and seam-based components across multiple manufacturing sites. Every new line or variant starts with a Process Engineer working out which machines, tooling and poka-yoke devices are allowed for a given part.

Historically that was done by cross-referencing spreadsheets, emailed specs and direct entries into the PLM system.

What a configuration had to reconcile
Approved machines
Tooling
Poka-yoke devices
Site parameters
Seam type
Product family
The challenge

Errors that only surface after the fact

Because nothing enforced which machine, tool or poka-yoke combination was valid for a given seam or product family, invalid configurations were not caught until much later.

Gap one

No enforcement

Master data existed, but nothing stopped an engineer from selecting a machine or tool that was not actually approved for that site or seam type.

Gap two

No traceability

Configuration attempts left no audit trail. There was no record of who changed what, when, or why a version was superseded.

Gap three

No feedback loop

When master data was missing or inconsistent, engineers had no structured way to flag the gap back to Central Engineering and Purchasing, so they worked around it instead.

Business drivers

Why it was worth fixing

Reduce rework, standardise across plants, and give leadership real-time visibility into price, floor space and headcount from quotation through serial production.

Every workaround an engineer invents to get past a missing rule is a rule the next engineer will not know to follow.

The mandate

Four jobs the tool must do

  1. Configure

    Let a Process Engineer build a valid line configuration, with its seam table, operations, machines and tooling, using nothing but approved master data.

  2. Validate

    Block invalid combinations at the point of selection, not after the fact. If a machine or tool is superseded, or does not match the site or seam, the engineer should not be able to pick it.

  3. Balance

    Give engineers a way to distribute cycle time across operations and see, before committing, whether the line balances against the target tact time.

  4. Prove

    Keep a full version history and audit trail, covering every quotation, every master-data change and every action, so a configuration can be defended after the fact.

The solution

A five-step quotation wizard, gated by validation

Each tab feeds the next, and nothing generates until every mandatory field across all five is filled and valid.

  1. Request and demand

    Site, part, product family and demand volumes are captured. Site and product family are pulled straight from master data, never free-typed.

  2. Seam table

    The engineer enters seam-level parameters from the drawing: stitch type, needle count, thread cut. This becomes the filter for the next step.

  3. Operations

    Each operation picks a machine, tooling and poka-yoke device, but only from options pre-filtered by site, seam data and status. Nothing superseded is selectable.

  4. Balance and planning

    Cycle time, operator count, tact time and utilisation are calculated live. Engineers can borrow time between operations to balance the line before committing.

  5. Preview and save

    Every prior tab's data is shown on one page for a final check, then either saved as an editable draft or generated. Once generated it is locked, read-only, and visible to every Process Engineer.

Where a quotation lives

A folder tree, five levels deep

Site, then product family, then product name, then part number, then the quotation itself, with draft and archived states, cloning, PDF export, and a set as serial flag to mark the active configuration.

The gate

Validation happens at selection, not at save

The filter is applied to the options list itself, so an invalid machine, tool or device is never offered. An engineer cannot build an invalid configuration and discover it on submit.

Architecture

Six zones, layered for security and auditability

Isolated services behind an API gateway, with every zone able to reach only the one below it. Nothing talks to the data layer directly except the services that own it.

Frontend
Where an engineer's request enters. The list page, the five-tab wizard, the master-data and log views.
Identity
The request is authenticated against Azure AD before it reaches anything else.
API gateway
The single entry point to the services behind it. No zone skips the one in front of it.
Builder API service
Owns the configuration logic and the validation rules, and is the only service that writes to the register.
Data and evidence
PostgreSQL for the configuration data and Blob Storage for attachments and exports.
Events
The service emits on RabbitMQ, and the notification and audit log services pick that up independently of each other.
End to end

An engineer's request enters through the frontend, is authenticated against Azure AD, passes through the API gateway into the Builder API service, which persists to PostgreSQL and Blob Storage and emits an event on RabbitMQ. The notification and audit log services consume that event independently, so the engineer's action and its downstream side-effects are decoupled.

Inside the build

Five decisions that shaped what the tool can be trusted to do

  1. Keep seam input manual, for now

    The obvious next step is pulling seam parameters straight from the drawing. That was deliberately descoped for the MVP.

    Manual entry is slower, but it ships now, and it keeps the validation logic decoupled from a drawing-parsing integration that would have doubled the surface area to test.

  2. Operation numbers in multiples of ten, never renumbered

    Operations default to 10, 20, 30. When an engineer needs to insert a step between two existing ones, they type a number in between, say 14, rather than triggering a renumber of everything downstream.

    No auto-renumbering means no silent reshuffling of a sequence someone else is relying on.

  3. Line balancing by borrowing, not by rebalancing everything

    Rather than an algorithm that redistributes cycle time automatically, an engineer manually chooses which operation lends time to another, and by how much.

    It is slower than automation, but every rebalancing decision is visible and attributable, which matters more for a first release the team has to trust.

  4. Generated quotations are locked and public, drafts are private

    A draft is visible only to its author and stays fully editable. The moment it is generated it becomes read-only and visible to every Process Engineer.

    That trades flexibility for the guarantee that a generated number never silently changes underneath a decision already made on it.

  5. Cloning copies data, not attachments

    Cloning a quotation carries forward every field except file attachments.

    A cloned quote is a new starting point, not a certified copy of the original evidence, so it cannot be mistaken for one.

Validation plan

Every performance test has a pass line, written before the test runs

Because this replaces a process people already distrust, the plan is to prove stability and correctness under load before go-live, not after.

The tests, and what each one proves
Concurrency and scale
Stability at 50 concurrent users over 20 minutesPass line: zero crashes, error rate under 1%
Master-data CRUD
Reads stay fast while Central Engineering edits master dataPass line: P95 at or under 2.0 seconds, read degradation under 15%
Auth and session burst
The Azure AD path under login spikesPass line: login success at or above 99%, no 401 spikes
Soak
No leaks over two to four hours of mixed usePass line: latency drift under 10%, no memory leak
Data integrity under load
Revisions and audit logs stay consistentPass line: 100% of sampled writes consistent
Health and monitoring
Failures are detected and routedPass line: alerts fire, mean time to acknowledge at or under 10 minutes
The non-functional targets these tests are written against
Target
50

concurrent users

Target
Under 2s

core-operation response

Target
99.9%

availability SLA

Target
1 year

log retention

These targets are defined in the Non-Functional Requirements and the Performance Test Plan. Results will be published here once testing completes.

In the room

What a Process Engineer sees

The list page
Opens to a folder tree, with On going and Archived tabs, searchable by quotation name and filterable to My Site.
Each quotation
Shows its status and a rename option, with a three-dot menu carrying archive, download, clone, and set as the active serial version.
Every category has its own log
Central Engineering master data, pricing and quotations each carry a system log: who changed what, from which value to which, and when.
The log is a working surface
Searchable, sortable and exportable to CSV, rather than a write-only record nobody can interrogate.
Targeted outcomes

What success is defined as, and what has not been measured yet

These are the numbers the Solution Vision commits to. None of them is measured, because the tool has not gone live. They are published here so this page can be honestly updated against them.

Target
30%+

rework reduction

Target
20 to 25%

shorter lead time

Target
Over 90%

first time right

Target
One

shared master-data source

Claim by claim, with its status

Rework reduced by 30% or moreTo be measured post go-live

Target

Lead time cut by 20 to 25%To be measured post go-live

Target

Over 90% first time rightTo be measured post go-live

Target

Non-functional performance targetsTest plan defined, execution pending

Planned

A case study that reports a target as a result is worth less than one that reports nothing, because the next reader has no way to tell which of its numbers were earned.

What comes next

Deliberately left for later

The MVP drew a hard line around what it would and would not attempt. What sits outside that line, in order of likely priority.

Drawing-based seam extraction
Pull seam parameters directly from the drawing files instead of manual entry. This is decision A, revisited once the validation logic is proven on its own.
Automated line balancing
Replace manual borrowing with an optimizer that balances cycle time across stations. This is decision C, revisited once engineers trust the manual path.
Full PLM integration
Two-way sync, instead of the current manual bridge.
CAPEX approval flows
Route generated quotations through spend-based approval automatically.
KPI dashboards
Roll utilisation, cost and cycle-time data up into a management view.
In-tool engineering notifications
Alert Central Engineering automatically when a station flags red or yellow, instead of relying on manual follow-up. This closes the missing feedback loop from section 02.
Next step

Which of your rules only exist in a document?

If the approved list lives in a spreadsheet and the enforcement lives in somebody's memory, the gap shows up in procurement or on the floor. We can walk one of your configuration paths end to end in a single session and show you where the rule stops being enforced.