Changeover Matrix

What this tool does

CM8-355 holds the changeover time between every pair of products on a line, in both directions, and shows you the whole picture at once: which pairs are expensive, which are far worse one way round than the other, how much of the time is spent waiting for approval rather than working, and what a sensible running order would save.

Everything runs inside this single file — no account, no upload, no network request of any kind.

What one row is

One row is one changeover in one direction on one line. Clear to blue and blue to clear are two rows. The same pair on two different lines is two rows.

You do not need every pair before the tool is useful. Start with the pairs you actually run, and the matrix will show gaps as a dot — which is itself informative, because a blank cell is usually either a sequence you never run, or one nobody has ever timed.

Direction matters

This is the single most important idea in the tool, and it is the one most changeover records get wrong by holding a single time per product.

Changeovers are frequently asymmetric, often dramatically. Light to dark needs no cleaning; dark to light needs a full flush. Fine to coarse is a rinse; coarse to fine is a strip-down. Small to large may need no change part; large to small does. A record that stores one number per product cannot see any of this, and the sequencing decisions made from it are close to random.

Defining the changeover time

Use one definition and use it everywhere:

Not "when the fitters finished". Not "when the line restarted". The clock stops when the line is making sellable product again, which means first-off approval and any settling-in are inside the time, not after it. This definition is stricter than most people expect and it is the one that matches what the changeover actually costs you.

Reading the matrix

The matrix table is the centrepiece. Rows are what you are changing from, columns what you are changing to, and each cell is the time in minutes. A cell marked ! is over your target; a dot means that pair has never been recorded; the diagonal is dashes because a product does not change over to itself.

Read it by looking for rows that are uniformly expensive — a product that is costly to leave, whatever you go to next — and columns that are uniformly expensive, a product that is costly to arrive at. In the sample, clear is cheap to leave and expensive to arrive at, because every return to clear needs a flush. That single observation is worth more than the average of anything.

The matrix shows up to ten products so it stays readable on a printed page; the rest of the tool uses every row you have entered.

Asymmetry, and why it is free money

The tool finds the reverse of each pair automatically, matching on the same line. Where a pair is badly one-sided, the running order alone decides which number you pay, and changing a running order costs nothing. No engineering, no capital, no kaizen event — just doing the same work in a different sequence.

This is almost always the first saving available on a line, and it is almost always invisible until somebody writes both directions down.

The running order, and what it is not

The second table proposes a sequence visiting every product once, built by a nearest-neighbour search: start somewhere, always go to the cheapest unvisited product next, repeat from every possible starting product and keep the best result.

Two honest caveats, both of which matter:

• It is a heuristic, not an optimum. Finding a provably best order is the travelling salesman problem. For a handful of products nearest-neighbour is usually optimal or within a few percent; for twenty it may not be. It will never be worse than the order you fell into by accident.
• It knows nothing about your schedule. Due dates, batch sizes, stock cover, shelf life, campaign lengths and what the customer wants on Thursday are all absent. This is the sequence that minimises changeover time and nothing else, which makes it an input to scheduling and not a schedule.

Where a pair has never been recorded, the search uses the average of everything you have recorded, and the table says so in the Basis column. Treat a proposed order that leans on several unrecorded pairs as a prompt to go and time them.

Where the saving figure comes from

The comparison is against a random order, not against what you do today. That is deliberate and conservative in one direction and optimistic in the other: if your current sequence was chosen with any thought at all, the real saving is smaller; if it is genuinely arbitrary, this is about right. To find the true figure, add up your actual current sequence by hand from the matrix and compare — that takes two minutes and is the number worth quoting.

First-off approval

Record how much of each changeover is spent waiting for the first piece to be approved. It is part of the changeover time, not extra, and the tool checks that you have not entered it as more than the whole.

It gets its own tile and its own segment in the driver chart because it is usually the largest single component that requires no engineering to remove. The fixes are organisational: have the inspector present at the changeover rather than called afterwards, let the operator sign off within defined limits, pre-set what "good" looks like so the judgement takes a minute rather than twenty. Lines routinely find a fifth of their changeover time sitting in this column.

Costing a changeover

Use contribution lost per minute if the line is the constraint and you could sell everything you make; use the cost of running the line if it is not. The two can differ by a factor of five, so say which you used when you present the figures.

Scrap through the changeover is included because for wet processes it is often the larger half. In the sample, the flush changeovers scrap five times as much product as the ones that need no flush.

This is not SMED

SMED — separating internal from external work, converting internal to external, streamlining what remains — is how you make an individual changeover faster. This tool does something different and complementary: it tells you which changeovers are worth attacking, and it collects the saving available from sequence before anybody touches the machine.

Use the matrix to choose the target, then use a SMED study on the pair that costs the most.

Limits worth knowing

• It records one time per pair. Real changeovers vary, and a single figure hides that variation — if a pair swings between twenty and ninety minutes, the average is the least useful thing about it.
• Rows marked "estimate" are placeholders. They are flagged in the Position column and in the chart labels so an estimate never quietly becomes a fact.
• Labour minutes are recorded but not costed, because in most plants the constraint is the line rather than the people. If that is not true for you, cost the labour separately.
• It handles one line at a time well and several only if you use the line field consistently — the matrix table does not split by line.

Printing and sharing

The Report tab prints the tiles, charts, the matrix and the running order with a title block you fill in. The matrix on its own, printed and pinned up next to the schedule board, is the highest-value page this tool produces.

Saving your work

The matrix is held in this browser, on this computer, and stays there between visits. Use the backup button to write a JSON file you control; the spreadsheet download gives you the same rows.

Accuracy & disclaimer

Every time in this tool was measured or estimated by you. The tool sorts, subtracts and searches; it has no view about whether a changeover could be faster, whether a flush is necessary, or whether a sequence is safe to run. Nothing here should override a process instruction, a cleaning validation or an allergen control.