Stationline user guide

Getting started

Stationline keeps one project with four views that share their numbers.

1. Line balance. Enter the available time per period and the demand; the tool divides them to get the takt time. List the tasks, each with a time in seconds and the tasks that must come before it. Stationline assigns tasks to stations so that no station is busier than the cycle time and no task runs before its predecessors.
2. Layout. Give the departments, a from-to chart of the loads moved between them and, if you like, a relationship chart. Stationline finds a good grid layout and scores yours.
3. Value stream. Describe the process steps in order, with the piles of work waiting in front of each. Stationline gives lead time, process time and process cycle efficiency from the same takt time as the line.
4. Work measurement. Enter stopwatch readings with a performance rating and an allowance to get normal and standard times, plan work sampling, or project a learning curve. A task in the line balance can take its time from a study, and a value-stream step can take its cycle time from a balanced station, so a change in one view flows into the others.

Use Examples for a worked starting point and Save to keep the project as an open JSON file; Open reads it back. Your work is also autosaved in this browser. Nothing is uploaded.

A worked example you can check by hand: seven tasks of 2, 7, 4, 3, 6, 5 and 3 s, with task 1 and task 2 before 3, task 2 before 4, tasks 3 and 4 before 5, task 4 before 6, and tasks 5 and 6 before 7, and a 10 s cycle time. The 30 s of work cannot fit in fewer than three stations; the proven minimum is four. Ranked positional weight uses four stations with loads 10, 6, 6 and 8 s, so efficiency is 30 / 40 = 75 %.

Line balancing: takt, cycle time and the rules

Takt time is the pace of demand: available time per period divided by the demand per period. Both numbers must describe the same period, and available time should be the time the line really runs after planned breaks. The cycle time is the longest time any station may take. By default it equals the takt time; you may enter a shorter cycle time to leave a margin, or a longer one to see the effect, in which case the tool says that the line cannot meet demand. A task longer than the cycle time cannot be placed, so the tool stops with a message that names it instead of guessing.

The three priority rules fill one station at a time. Among the tasks whose predecessors are all placed and that still fit in the open station, they take:

• Largest candidate rule - the one with the longest time;
• Ranked positional weight - the one with the largest positional weight, which is its own time plus the time of every task that must come after it, directly or indirectly;
• Most following tasks - the one with the most tasks that must follow it.

Ties are broken by the longer task, then by list order. When nothing fits, a new station opens. These are quick rules of thumb: they can use more stations than the minimum.

The exact check finds the proven minimum number of stations for up to 30 tasks by working through every combination of tasks that respects the precedence relations. It refuses, with a message, if that needs more than 500,000 partial states. The comparison table shows how many stations each rule is above the minimum.

Limits: up to 500 tasks, times from 0 to 1,000,000 s. Zero-time tasks are allowed. This is a single-model, paced, straight line with fixed task times and one worker group per station; it does not do mixed models, parallel stations, U-shaped lines, zoning constraints, setup times or worker skills.

Efficiency, balance delay and smoothness

With S stations, cycle time C and total work content T (the sum of all task times):

• Efficiency = T / (S × C). The share of the paid station time that is used for work.
• Balance delay = 1 − efficiency. The share that is idle.
• Idle time per cycle = S × C − T.
• Smoothness index = square root of the sum, over stations, of (longest station load − station load) squared. Zero means perfectly even stations; larger means more uneven.
• Theoretical minimum stations = T / C rounded up. No balance can use fewer; the exact check shows when more are really needed because of the precedence relations.
• Critical path = the longest chain of tasks. A unit cannot be finished faster than this, whatever the balance.

All of these are computed once per balance and shown identically in the tiles, the comparison table, the explanation and the exports.

Layout: from-to chart and adjacency

The plant is a grid of equal cells, rows × columns, with a cell pitch in metres. Each department takes one cell. Distances are measured between cell centres, either rectilinear (along aisles) or straight line, multiplied by the pitch.

Flow objective. The from-to chart gives the loads moved from each department (row) to each other department (column) per period. Cost = sum over all ordered pairs of flow × distance × cost per load per metre. The chart need not be symmetric. The tool minimises this cost.

Adjacency objective. The relationship chart gives each pair one letter: A (absolutely necessary), E (especially important), I (important), O (ordinary), U (unimportant), X (undesirable). Each letter has a numeric weight you can change; the defaults are A 4, E 3, I 2, O 1, U 0 and X −4 and are teaching values, not a standard. The score is the sum of the weights of the pairs whose cells share an edge. The tool maximises it. Diagonal neighbours are not adjacent.

Up to 9 cells the tool checks every arrangement, so the layout it proposes is best possible (ties are possible, and one best layout is shown). Beyond that it uses a pairwise-exchange improvement search from several fixed starting layouts: a good layout, labelled as a search result, never as proven best. Both scores are always shown for both layouts, and the lists of A or E pairs that are apart and X pairs that touch refer to the layout named in their heading.

Limits: up to 30 departments on a grid of at most 12 × 12 cells. All departments have the same size; unequal areas, fixed walls, doors and aisle widths are not modelled.

Value stream: lead time and process cycle efficiency

A serial value stream lists the steps in order. For each step enter the cycle time per piece, the changeover time and how many pieces are made per changeover, the uptime (the fraction of time the step is available) and the pieces waiting in front of it. Then:

• Takt time is the same as on the line balance.
• Effective cycle = cycle time / uptime + changeover / pieces per changeover. A step whose effective cycle exceeds takt is flagged as too slow.
• Capacity per period = available time / effective cycle.
• Waiting time of a pile = pieces × takt time. This is Little's law: the wait equals the pieces in the system divided by the rate at which pieces leave. The stream total includes the finished pieces after the last step.
• Process time = sum of the cycle times. Lead time = waiting time + process time.
• Process cycle efficiency = process time / lead time.
• Operators for one piece per takt = process time / takt time. The operators you enter per step are added up and compared with this minimum; they do not change the cycle times, which you enter or take from a balanced station.

Times are available production seconds, not calendar time. A step can take its cycle time from a balanced station of the line view, so the stream follows the balance. This is a basic metric calculation for one product family; it does not draw information flow, handle several product families or simulate variation.

Time study, work sampling and learning curves

Time study. For each work element enter the observed times, a performance rating (100 % is a normal pace) and how many times the element occurs per work cycle. Then:

• Normal time = mean observed time × rating / 100 × occurrences.
• Standard time adds the allowance for personal needs, fatigue and delays. If the allowance A is a share of normal time: standard = normal × (1 + A). If it is a share of the whole shift: standard = normal / (1 − A).
• Observations needed = (t × s / (precision × mean))² rounded up, where s is the sample standard deviation and t is the Student t value for your confidence and n − 1 degrees of freedom. It is never below the readings you already have. With one reading no spread exists and no figure is given.

You must supply your own ratings and allowances: Stationline contains no predetermined-motion-time tables or published allowance tables, and the ratings in the examples are invented.

Work sampling. With n random observations of which m show the activity, the proportion is m / n with half-width z × √(p(1 − p) / n) at your confidence (normal approximation, clipped to 0–100 %). The observations needed for a half-width e are z² p (1 − p) / e²; a second figure uses the most cautious p = 0.5.

Learning curve. The Wright model gives the time of unit n as T₁ × n^b with b = log₂(rate): at an 80 % rate each doubling of output multiplies the unit time by 0.8. The total for a range of units is the exact sum of the unit times (up to 1,000,000 units).

Limits: up to 20 studies, 40 elements per study, 200 readings per element.

Limitations: what Stationline does not do

• Line balancing is single-model, paced and straight, with fixed task times: no mixed-model sequencing, parallel or U-shaped stations, zoning, setup or learning effects, worker skills or buffers, and no stochastic task times. At most 500 tasks; the exact check at most 30.
• Layout treats every department as one equal cell on a rectangular grid: no unequal areas, walls, doors, aisle widths, multi-floor plants or material-handling equipment limits. Beyond 9 cells the result is a search result, not a proof.
• Value stream is a serial, single-product metric calculation with deterministic times: no information flow, several product families, variation, scrap or rework.
• Work measurement uses your own ratings and allowances. There are no predetermined-time systems, rating tables or allowance tables, and results are not a labour standard or a basis for pay.
• Diagrams stay legible up to about 60 tasks, 30 departments and 40 stream steps. Larger precedence diagrams are drawn without labels and wrap into bands of 12 levels; use the tables for the numbers.
• Nothing here has been audited or approved by any body, and nothing replaces the judgement of a qualified professional.

Files, units and privacy

All times are in seconds, counts in pieces, and layout distances in metres; there is no hidden unit conversion. The project file is open JSON with a schema version; older files are upgraded when opened, and a file from a newer version is refused with a message. Unknown fields are refused rather than ignored.

CSV tables can be imported and exported: tasks (id, name, time_s, predecessors separated by semicolons), the from-to chart, the readings of a study and the station assignment. Exported CSV starts with a marker line and text cells that could be mistaken for spreadsheet formulas are protected with a leading apostrophe; the same marker tells Stationline to remove it when the file is read back. Drawings export as SVG or PNG from the Export menu and always show the model as last calculated: while a calculation is running the export buttons are disabled.

Files are limited to 1 MiB and 5,000 CSV rows. Nothing you enter leaves your browser except a signed token request that lets the calculation worker start.

How we verify

The Verify page runs 35 cases (177 individual comparisons) in 14 distinct validations, in your browser, with the same worker that does your calculations. A family of related cases counts once.

• Independent code. Seeded random instances are solved by a separate Python program: it re-implements the priority rules as a ranked list, finds the minimum by exhaustive enumeration, enumerates every layout, and uses SciPy and exact fractions for the statistics and the stream. Student t values also match a government statistical handbook table.
• Hand derivations. Station-by-station traces of each rule on the seven-task graph, metric identities and the cross-view links are worked by hand in the case text, so they check the code against the formulas as written, not against a second source.
• Properties. The automated test suite also checks, on hundreds of seeded random models, that every assignment respects capacity and precedence, that efficiency times capacity equals the work content, that scaling all times leaves the assignment unchanged, and that the exact check equals brute force.
• Published results (development tests only). The project's own test suite, not this Verify page, also compares the exact check with proven minimum station counts of published benchmark graphs (Scholl 1993; Scholl and Klein 1997) and the layout search with published optimal costs of the Nugent layout problems (QAPLIB). Those data are kept out of the shipped tool.

No published worked example of a complete value-stream map is used: its numbers are checked by hand and by independent code, not against a published case. The heuristic rules are checked against an independent implementation, not against published step-by-step results.

Passing these checks shows the arithmetic and search procedures are implemented as described. It does not certify a production line, a layout or a labour standard.

Compared with other tools

Stationline is free, needs no installation and keeps your data in the browser. Compared with spreadsheets, it enforces the precedence relations, tries three priority rules at once, proves the minimum station count for small lines, and keeps takt, task times and stream cycle times consistent across views. Compared with dedicated commercial packages for line balancing, layout and work measurement it is deliberately small: it has no motion-time databases, no simulation of variation, no mixed-model sequencing, no drawing of building plans and no multi-user features. Use those tools where you need them; use this one to learn the methods, to check a small case or to explore a decision before investing in more.

Sources

The methods are standard textbook methods; the formulas are written out in full in the formula register that ships with the tool, with citations to: A. Scholl, Balancing and Sequencing of Assembly Lines (Physica, 1999); F. B. Talbot, J. H. Patterson and W. V. Gehrlein, A comparative evaluation of heuristic line balancing techniques, Management Science 32 (1986); W. B. Helgeson and D. P. Birnie, Assembly line balancing using the ranked positional weight technique, Journal of Industrial Engineering 12 (1961); C. E. Nugent, T. E. Vollmann and J. Ruml, Operations Research 16 (1968); R. E. Burkard, S. E. Karisch and F. Rendl, QAPLIB, Journal of Global Optimization 10 (1997); R. Muther, Systematic Layout Planning (1973); J. D. C. Little, Operations Research 9 (1961); T. P. Wright, Journal of the Aeronautical Sciences 3 (1936); and the NIST/SEMATECH e-Handbook of Statistical Methods.

Results are for education and preliminary planning. They must be checked by a qualified professional and are not a substitute for a measured labour standard, an agreement with workers or licensed engineering judgement.