Manufacturing / Bottleneck detection

Production bottleneck detection from the data your plant already has

For aerospace and industrial manufacturers building complex assemblies. Mirage reads work in progress from your ERP production orders, MES or shop-floor exports, spreadsheets and planners' emails, compares each order with its routing and dates, and flags the station whose queue is growing or the component holding up an assembly, with the evidence, for the person who decides.

No bottleneck detection deployment is published yet. The exception layer it builds on runs in Transwin's control room.

The reality

Work in progress is spread over the ERP, an MES or shop-floor export, a planner's spreadsheet and a thread of emails. Each one says something true, and none says where the work is piling up. A late assembly is usually found when it is due, by someone walking the floor or chasing a part, not on the day the queue in front of one station started to grow.

What Mirage does

How production bottleneck detection works

Mirage sits beside your systems, not inside them. It reads what they already record, rebuilds where each order stands against its plan and raises what has drifted.

Reads WIP where it already lives

Your ERP production orders and their operations, MES or shop-floor exports, the planners' spreadsheets and the emails that move dates. Each order is matched to its routing and to the components its assembly needs. Nothing on the floor is replaced, and Mirage can run in your infrastructure or your own cloud account, so the data stays in the environment you control.

Compares each order with its plan

Per operation: started, finished or waiting, and how long it has waited against its planned date. Per station: how many orders wait in front of it, and whether that number grows from one export to the next. Per assembly: which component is not there yet, and which orders it holds up.

Alerts with the evidence, a person decides

An alert names the station or the component, the orders affected, the records it was built from and the planned date each one is missing. The planner or the production manager decides the action: resequence, expedite, add a shift. The agent changes no order and no schedule by itself.

Bottleneck, queue, WIP: what the agent measures

The terms production teams already use, stated plainly. No proprietary score.

A bottleneck is the step that limits output

The station, machine or team whose capacity sets the pace of the whole flow. Work arriving faster than it can process waits in front of it, which is why a growing queue is the first visible sign. The Theory of Constraints, credited to Eliyahu Goldratt, starts by identifying that constraint, then exploits it, subordinates the rest of the flow to it and elevates it (Theory of Constraints Institute, five focusing steps).

Queue and waiting time

The orders waiting in front of a station, and how long the oldest has waited compared with its planned start. A queue that is long but stable is a different signal from one that grows every day, and the agent reports both the count and its trend.

WIP and cycle time move together

Over a stable period, average work in progress equals throughput multiplied by the average time an order spends in the system: Little's law, proved by John Little in Operations Research in 1961. When WIP rises while output stays flat, orders are spending longer on the floor.

A late component is a bottleneck too

In a complex assembly, one missing part stops the final operation while every other component waits. The agent follows the components listed on each assembly order to name the part, the order it is late on and the assemblies it holds up.

The output

WIP by station, as the agent reports it

An illustrative report for one day on an assembly line. Every value is invented.

StationOrders waitingOldest order against planStatus
Machining cell 24On its planned dateOn plan
Heat treatment↳ Eleven orders wait in front of heat treatment, up from six in last Monday's export, and the oldest has waited three days past its planned start. The alert lists the eleven orders and the export rows behind each count.113 days past its planned startQueue growing
Sub-assembly B6Same day as plannedOn plan
Final assembly↳ Two assembly orders are complete except for one bracket, late on its purchase order. The agent does not choose between expediting and resequencing: the alert goes to the planner with both orders and the supplier's last confirmed date.25 days late, waiting for bracket P-2210Held for a person

Proof

What Mirage has in production, and what it has not

Mirage has not published a production bottleneck detection deployment, and has no manufacturing WIP client to name. The nearest published work is the event and exception layer of its operations control rooms, and ERP order entry at Biodéal. Each card says what it is.

Transwin (operations control room, logistics)

1 control room across the business, 5 systems in production

Surfaces the anomalies and events that need a human

Freight, fuel, mining and fleet, not a production line

See the case study

Mass Céréales Al Maghreb (control room being deployed)

One control layer being deployed across 2 port terminals

Exceptions and alerts in one operational view

No result figure published yet

See the case study

Biodéal (order entry on SAP Business One)

48 seconds median from email received to order ready in the ERP

About 7% of orders flagged for a person to check

Order entry into the ERP, not WIP tracking

See the case study

Production bottleneck detection, on a week of your data

Send us a week of WIP exports: production orders with their operations and dates, an MES or shop-floor extract, the planners' spreadsheet. We come back with the WIP by station as the agent would report it, the queues and late components it would flag, and what reading your systems directly would involve.

Attach WIP exports (recommended)

We use your email to reply to you and, now and then, to tell you about Mirage. You can opt out at any time. Privacy policy

How a deployment runs: Observe, Map, Build, Run, Expand. See the method

FAQ

Frequently asked questions

  • Look for where work waits. Compare each order with its routing and planned dates, count the orders waiting in front of each station and how long the oldest has waited, and follow the components each assembly is missing. A station whose queue keeps growing, or a part that holds up several assemblies, is where to look first. Mirage does this from the exports your plant already produces.

  • Your ERP production orders with their operations and planned dates, MES or shop-floor exports showing what was started and finished, and, where they exist, the planners' spreadsheets and the emails that move dates. A week of exports is enough for a first proof of concept. No new sensor or terminal on the floor.

  • Mirage has not published one, and has no manufacturing WIP client to name. The nearest published deployments are the event and exception layer of its operations control rooms: in production at Transwin, a logistics operator (one control room, five systems), and being deployed at Mass Céréales Al Maghreb's port terminals, with no figures yet. Its ERP order entry runs at Biodéal.

  • No. It reads and alerts. Each alert carries the station or the component, the orders affected and the records behind them, and the planner or the production manager decides the action: resequence, expedite, add a shift.

  • It supports its first step, identifying the constraint: the step that limits output. The Theory of Constraints, credited to Eliyahu Goldratt, then exploits the constraint, subordinates the rest of the flow to it and elevates it. Those are management decisions; the agent supplies the evidence for them and does not take them.

  • Mirage can run in your infrastructure or your own cloud account, where data and processing stay in the environment you control and the encryption keys are yours. Mirage's access is limited to what the deployment needs, and you can remove it.

  • A proof of concept on your own WIP exports one week after the first meeting, and the first workflow in production in 4 to 6 weeks. Reading several systems directly can take longer and is scoped at the first meeting.

  • A fixed-price pilot on a week of your WIP exports first, which you can stop after. Then an annual subscription priced by volume, quoted in the first conversation.