Reliafy
← What's new

What's new in Reliafy — November 2026

This month's three headlines all start from the data you already have: condition readings, your shutdown calendar, and the test you're about to plan.

Condition monitoring: each unit's remaining life, updated as readings come in

Most reliability models give every unit in a population the same risk at the same age. But you know more than its age. You have vibration readings, wear measurements, oil analysis, temperatures. Remaining life from condition readings puts the readings and the failure history together, so each unit gets its own risk of failure, and that risk updates every time a new reading arrives.

Give it two tables: one row per unit (its age, whether it failed, and anything else you know about it, such as its load or duty), and the readings taken over each unit's life. Reliafy fits how the readings change over time and how they drive failure. You get:

  • each running unit's risk of failing in the next 30, 90 and 365 days, and its median remaining life;
  • a unit view: its readings so far, where they're heading, and its updating survival curve;
  • what drives failure, in plain sentences, for example how much each extra unit of vibration raises the risk of failure.

Then use a unit's prediction as a block in a reliability block diagram: "how likely is this line to get through the next month, given what the sensors say about each pump right now?"

Find it under Modelling → Condition monitoring. It's part of Reliafy Pro.

Plan replacement around your shutdowns

The best replacement age is only useful if you can replace things at that age. In practice, most planned replacements happen at a shutdown. The new Shutdown calendar mode in Optimal replacement plans around the shutdowns you already have, each with its own cost: weekly stops, a six-monthly shut, the annual major.

  • The rule in plain words: for example, with a shut every 180 days, "at the shut, replace the part if it's older than 138 days". In that example the plan costs about 20% less than running to failure, and 11% less than replacing at a fixed age.
  • The cost a year compared with running to failure, plain age replacement, and replacing at every shut.
  • The next 12 months as a schedule, with real dates.
  • One unit right now: enter its age and see whether to replace it now, at the next shut, or leave it, with the break-even ages.

Exact answers are free on every plan.

A test planner, before you spend the test budget

The demonstration test calculator is now a full Test planning page under Strategy:

  • Demonstration: how many units, for how long, to show a reliability or an MTBF at a given confidence, now also for failure-terminated tests.
  • Sequential: accept or reject as the results come in, often with less testing than a fixed-length test.
  • Life test: how precisely a test will estimate a B-life or the mean, or how many units and how long you need for the precision you want.
  • Accelerated test: the best stress levels and how many units to put at each, or a check of a plan you have in mind. It opens pre-filled from any accelerated life model.
  • Comparison: will a test be able to tell whether a change worked? Units and power for comparing two groups, and the operating time you need before and after.

All the plans are free. Checking a plan by simulation is part of Pro.

The Reliafy API and command line now do everything your agent can

Everything your AI agent can do with Reliafy, a script can now do through the Reliafy API: fit, analyse, build and change diagrams, plan tests, forecast fleets. There's also a command-line tool:

pip install reliafy-client
reliafy login
reliafy fit failures.csv --dist weibull
reliafy tools

It's handy for scheduled jobs, such as pushing a fleet's usage every night, or for running an analysis over a whole folder of files. The API is part of Reliafy Pro.

Also new and improved

  • Life data: B-lives and MTTF with confidence bounds; requirement checks; competing risks, now with a 95% band for each failure mode; time-varying covariates; more regression models; recurrent-event models.
  • Block diagrams:
    • rename blocks and add them in place;
    • enter a block by its MTBF;
    • a starter for safety functions;
    • Kijima repair and mixture lives;
    • allocate a target across blocks;
    • production availability;
    • phased missions and networks;
    • time-varying conditions;
    • "What to improve" measures;
    • discounted costs;
    • exact Compare;
    • run history with time estimates.
  • Common-cause groups with proof tests are now included in the exact failure frequency and up and down times.
  • Proportional-hazards check: now the exact test used by R's cox.zph.
  • Reliability growth projections with BC modes use the bias-corrected shape, which matches the handbook's worked example.
  • Uncertainty for limited-failure and zero-inflated fits now includes the share that fails.
  • Additive-hazards models now show a blank outside the model's valid range, rather than a reliability above 1.
  • Compare groups now treats a blank failure mode as a unit still running, the same as fitting does. If you compared groups on data with a failure-mode column, it's worth running the comparison again.

As always, if something doesn't work the way you expect, reply to this email — it comes straight to me.

Derryn