Start hereNetwork

This page tells the story in order: first the problem, then how glass is made step by step, then what each screen is for, then where to start by job. If you have never seen a glass plant, read it from the top.

A ninety second walk through four ideas: the furnace does not wait for demand, losses are found at the level where someone can act, a job change follows the glass through the machine, and software detects while a person stays accountable. Pause at any step.

The whole business on one page: four plants, which of their bottle lines are running, how much glass each furnace melts and how much of it went down the chute to be melted again. Choose a plant to step inside it.

The long view of the furnaces: how far through its life each one is, which gets rebuilt in which year, and whether the network can cope with one being cold. A rebuild takes a furnace out of service, so the order matters.

The business on one page for someone with minutes to spare: decisions waiting on you, pack to pull (the share of melted glass that left as saleable bottles), the money spent on remelting glass, the furnace rebuild horizon and the next furnace decision.

Quality for all four plants side by side. It shows how many bottles each plant rejects, which faults are turning up at more than one plant, and whether the inspection cameras at different plants judge the same bottle the same way.

Faults that start in the melt, furnace by furnace. A blister is a trapped gas bubble and a streak is a ribbon of uneven glass, usually from the furnace lining wearing away. Both come from the furnace, so they hit every bottle line it feeds at once.

Every mould set the business owns. A mould set is the matched group of metal moulds that shapes one bottle design. This screen shows how worn each set is and which designs keep producing bad bottles.

What the plants buy to make glass: sand, soda ash and limestone, bought in recycled glass, gas and power. It shows what is on hand, what it costs in each state and which silos are running low.

Forming is the step that shapes molten glass into bottles. This screen compares how each bottle design runs on each machine across all plants, shows which pairings run badly and records what was tried.

Follows a customer complaint back through the pallet, the date code and the job to the shift and the exact mould position that made the bottle. It also ranks holds by what they cost.

Demand against days of stock for every bottle design across the network, and which lines are physically able to make each design. Diameter, weight and glass colour decide what a line can run.

Where finished bottles are stored across the four plants, how many days of demand that stock covers, and how much is on hold and so cannot ship.

Where the finished bottles are. Each plant's warehouse is drawn as a column filled with its pallets, every truck on the road is drawn where it is now, and a table shows which bottles run out in the next two weeks if nothing changes. Select a state to zoom in. Below that is what delivery costs: a truck fills its floor with pallets before it reaches its weight limit, so freight is priced by the pallet.

Which customers and bottle designs make money. For each it shows the cost of making and delivering the bottle and the carbon emitted, which depends on the electricity grid where the furnace stands.

A sustainability report draft. For each topic (climate, water, recycled glass, safety) it asks two questions: does the business harm or help people and the environment through it, and does it affect the business's own money. Each answer carries the numbers behind it and a link to the screen that owns them. A person checks it and signs the report; this screen cannot.

One connected view of the business, from a single line up to the whole network. It shows live signals, what the model has found and who owns it, with the numbers added up the way the plant is physically built.

Type a plain question about the plant and get an answer with the working shown. The question is matched to a fixed set of calculations on the same numbers as every other screen, and no language model is called.

A simple linear model, fitted live on the running plant, predicts what each line should be producing. The gap between that prediction and what the line actually produces is the useful part, because problems show up there first.

Standing checks that watch for conditions a person could miss. They are ordinary programs, not AI: each has a declared trigger, a limited view of the data and an audit record. They raise issues and propose actions, and a person decides.

The connections this platform would need to a plant's existing systems, designed but not connected. It states what this platform owns, what it only reads, which system is the master record for each field and what happens when one is down.

How the platform is built: where the assumptions live, how the plant is regenerated in your browser from a seed, and why two viewers always see the same numbers. There is no server and no database.

The plant drawn from above with live numbers on it, from the batch house where raw materials are mixed to the dock where trucks load. Click a furnace, a forehearth (the heated channel that carries glass to a machine) or a forming machine to open it.

The screen an operator watches in the control booth: live instrument readings, the targets they are held to and the alarm list. Colour appears only when something is abnormal, so a calm screen means a calm plant.

What happened on each shift, drafted from the shift's own data and handed to the next crew. A person signs it. It opens on safety, then covers output, quality and what the next crew needs to know.

One bottle line from end to end in a single drawing: the forming machine, the lehr (the long oven that cools bottles slowly), the inspection cameras and the pack centre behind it. Select any machine to see what it is doing and rejecting.

One section of a forming machine, the single station that makes a few bottles at a time. It shows the moulds, the gobs (the measured blobs of molten glass dropped into each mould), the cooling and the timing that explain why one bottle comes out wrong.

Recorded plant readings over time. By comparing a signal at the forming end with one at the inspection end, it recovers the delay between a cause and the moment the cold end sees it, and it declines to answer where the signal is too weak.

A register of what the model noticed, who owns each finding, what was done and whether the fix held. Software opens a finding. Only a person closes it, and that verdict is what the model learns from next.

A three week calendar of planned changes from one bottle design to another. It shows whether the mould set for each change is ready, the day each run becomes long enough to pay for its changeover, and when sets return from the shop.

The briefing before a line changes to a new bottle design. It carries forward the settings from the last run of that design, the actions its review left open and the mould set going on, and lists what must be ready before work starts.

The review done at the end of every run of one bottle design: what was lost against target, the three biggest faults and an action plan carried to the next run. Each run is judged against the band its changeover fell in.

The library of every bottle design this plant makes: which lines can make it, how many mould sets exist and where they are, and what each past run achieved. Each run shows pack to pull, the share of melted glass that left as saleable bottles.

The last part of the plant, where good bottles become pallets. It shows the conveyors feeding the machines that stack pallets, the wrapping stations, six driverless trucks, the warehouse lanes and the dock.

A directory of the departments that support the production lines, such as the store, the mould shop and forming support. Each has its own board, so choose the one whose work you want to see.

The workshop that repairs and prepares moulds. It shows which mould sets are on a bench, waiting, fitted to a machine or ready in the store, and how much life each has left.

The electricians' board: which lines are down because of a power, drive or control fault, the motors and controls on the cold end and pack centre, and what has been calling the department out. Anything on a forming section belongs to the machine department.

The board for the department that looks after the IS machine (individual section machine, the row of identical stations that shapes bottles). It lists every moving part past its service date, timing faults and stoppages in taking bottles out of the moulds.

The board for the trade that looks after the plant's own equipment: compressors, water, the batch house where raw materials are mixed, and every conveyor. It shows what is out, what is overdue for service and which bearings are warning of failure months ahead.

The compressed air, cooling air, water and material handling systems that keep the glass moving. It shows what each is asked for, what is running, how much spare capacity exists and what stops if one fails.

The front door to the three engineering departments: electrical, the forming machines and general plant maintenance. Work goes to the department that owns the equipment and the skill, so pick the one that matches the fault.

Every work order the plant has open or has closed, coded the way a reliability engineer reads them. It also shows each line's next planned stop, when the line is stopped on purpose so repairs can be done, and the mould sets and shutdown requests behind them.

A furnace melts glass day and night for ten to fifteen years and cannot be switched off. This screen shows how much each furnace in view melts, the energy it burns per tonne, how much recycled glass goes in and how far through its campaign (the years between rebuilds) it is.

The basement of the plant, where its own rejected bottles are crushed, sorted by colour and fed back into the furnace as cullet (crushed recycled glass). Each colour has its own conveyors and silos, because the colours can never be mixed.

The cold end is where bottles are inspected after cooling. This screen shows what the cameras and gauges reject, how much of that was good glass, and whether the inspection machines still agree with each other.

A hold stops a batch of bottles from shipping until someone decides what to do. Software raises a hold the moment a sample fails, and only a named person can release it. This screen lists every hold, how long it has waited and the tickets against it.

The order of runs on each line, and what a change to it costs. Switching a line to a different bottle takes hours, sends glass down the chute and moves days of stock, and this screen prices those changes before anyone commits.