For manufacturers
Forecast demand, materials, spare parts, and energy. Get an early warning when a machine starts drifting towards a failure. Free, open-source software, set up for you by the sktime developers.
No obligation. We look at your plant together.
How much should we produce when the season starts?
Is this bearing about to fail?
When will our plant hit its power peak tomorrow?
How many replacement seals should we keep in stock?
Illustrative example. sktime does the forecasting. The decisions stay with your team.
Real test-rig data: a bearing run until it failed (University of Cincinnati / NASA). sktime's CAPA detector compares all four bearings to their normal behaviour. Detector settings were chosen on this run, so read the lead time as an illustration. Data: IMS Bearing Data Set, NASA Prognostics Data Repository
Illustrative example. sktime does the forecasting. The decisions stay with your team.
Illustrative example. The forecast comes from sktime's Croston method, built for irregular demand. The range assumes random, independent demand.
Sound familiar?
A plant commits to many things long before they happen: shifts, purchases, spare parts, and power. These four decisions depend most on a good look ahead.
Production planning
Orders swing more the further they travel up the supply chain, a pattern known as the bullwhip effect. sktime forecasts demand per product and per week, with a likely range, so you plan with more than last year's numbers.
Materials
At the peak of the supply crisis, more than three quarters of German industrial firms reported material shortages. Before 2020 the average was 5.2%. sktime forecasts how much of each material you will need and when, so purchasing orders earlier and sees gaps before they reach the line.
Spare parts
Many spare parts are ordered rarely and irregularly. A literature review reports that in some industrial sectors up to 60% of spare parts show this pattern. sktime includes forecasting methods built for irregular demand, so you stock the right parts without tying up cash in the wrong ones.
Plant energy
For German industrial customers buying 2,000 to 20,000 MWh a year, the electricity price without taxes and levies rose from 7.3 cents per kWh in early 2021 to 18.1 cents in late 2024. A forecast of your plant's load helps you buy closer to what you use and see peaks coming.
Catch it early
A bearing, a pump, or a motor usually starts to behave a little differently before it stops: in vibration, temperature, or power draw. Your sensors already record this. Nobody can watch every signal around the clock, and a person sees the change only once it is large. sktime watches all signals together and reports when they leave their normal pattern. Published cases report warnings several days ahead for a bearing and two weeks to two months ahead for wind turbine parts.
sktime studies the sensor history from healthy operation. You do not need examples of failures. Failures are rare, so most plants have few records of them.
Live readings from all sensors of a machine are compared with that normal pattern. sktime reports the point where the behaviour starts to drift.
Maintenance receives an alert with the machine and the moment the change began. The team decides whether to order the part, plan the swap for the next stop, or take a closer look.
A bearing on a research test rig at the University of Cincinnati ran for about seven days until it failed. Its vibration was recorded every 10 minutes. The recordings of all four bearings went to sktime's detector, and we compared its warning with a simple alarm limit.
sktime early warning
3 daysbefore the failure. At that point, vibration was only about 10% above normal.
Simple alarm limit at twice normal
2 daysbefore the failure. A fixed limit like this fires only once the change is large.
False alarms in normal operation
0during the first 3.5 days, before the bearing started to drift.
One bearing on a test rig, with detector settings chosen on this run. Real plants differ: lead times depend on the type of failure, and we tune alert thresholds with your maintenance team on your own history.
Further reading:Test data: IMS Bearing Data Set, NASAFailures are rare: Zhu et al., surveyWind turbines, 2 weeks to 2 months ahead: Gigoni et al., 2019Bearing warning several days ahead: Seeq
Who it is for
Managing directors and plant managers
Better forecasts show what the plant has to deliver and what it needs to do so. Early warnings turn some unplanned stops into planned ones. You see the risk before it reaches the shop floor.
Maintenance, planning, and purchasing leads
You get one list of machines that need a closer look and a forecast of the parts and materials the next weeks will use. That lets you order earlier and schedule work in planned stops.
What the numbers say
Surveys and reports show what unplanned stops cost, what predictive maintenance can save, and how early most manufacturers still are with AI. sktime offers many forecasting and detection methods in one open-source toolkit, so you can use them without building everything from scratch.
$125,000
ABB surveyed 3,215 plant maintenance decision makers in 2023. Over two thirds of the businesses had an unplanned outage at least once a month, and 21% still repair machines only after they fail. Costs differ a lot by sector: Siemens reports $36,000 per hour in consumer goods and $2.3 million in automotive, in large plants.
8-12%
The US Department of Energy's maintenance guide cites studies of 8% to 12% savings over a program of preventive maintenance alone, and more for facilities that mostly repair after a failure. Deloitte reports 5% to 15% less downtime from its client projects.
24%
That leaves about three in four manufacturers with 10 or more employees at the start. Eurostat counts any AI use, including text and image recognition, so the share for forecasting is likely lower.
These figures come from surveys and reports by vendors, consultancies, and statistical offices, not from sktime projects. Results depend on your plant, your data, and how you maintain today. We work out what is realistic for your business in a first conversation.
sktime is open source: no license fees per user or site.
Your numbers
A back-of-the-envelope estimate. Enter what an hour of unplanned downtime costs you and how many hours you lose per year, and see a rough range in euros. The rest are our cautious assumptions, based on public reports.
–/ year
– less unplanned downtime per year (expected)
Not included: savings from better forecasts of demand, materials, and energy, and the cost of setup and support. After a first conversation we give you a fixed price for a trial project.
Three steps. The hours and the cost per hour come from you. The percentages are fixed assumptions, based on the sources below.
1. What unplanned downtime costs you
hours of unplanned downtime × cost per hour
You set both values. For scale: ABB found a typical cost close to $125,000 per hour in a 2023 survey of 3,215 plants. Siemens reports $36,000 per hour in consumer goods and $2.3 million in automotive, in large plants. At the start values of 150 hours and €50,000, this is €7.5 million a year.
2. What earlier warnings avoid
cost of downtime × 5 to 15% less
Deloitte reports 5% to 15% less facility downtime, based on its client work. At the Italian rail operator Trenitalia, downtime fell by an estimated 5% to 8%. We use 5% (cautious) to 15% (optimistic) as the range.
3. Annual savings
cost of downtime × the percentage from step 2
The expected value is the midpoint of the cautious and the optimistic case.
Annual savings = step 3.
Simplifications: we apply the percentages to all unplanned downtime, although not every failure gives a warning, and every hour costs the same.
This is an illustration, and we do not promise these results. The reduction rates are based on reports on other companies, not on sktime projects. Downtime costs vary a lot between sectors and plants. In the call we will explore options for your own use case.
Before you ask
How we start
We look at your data and at where bad forecasts or unplanned stops cost you the most today. From that, we set the scope and a fixed price for the pilot.
Over two to six months, we build a first system on your own data, for example a demand forecast or an early warning for one machine type, and compare it with how you work today. You get benchmarks, all the code, and a report, and decide from there how to continue.
Forecasts and alerts go into your daily planning and maintenance routines. Under a service agreement, sktime Enterprise Solutions maintains the system and stays your contact when questions come up.
Your first step
sktime is free and open source. In a free 60-minute call, the people behind sktime look at how you plan production and maintenance today, and what better forecasts and early warnings could be worth for your business.
sktime is downloaded over a million times a month and built by more than 500 contributors worldwide. Its core team now sets it up for you.
No obligation, no sales pressure.
What happens in the call
How you plan production, materials, and maintenance today, and where surprises cost you the most.
What you already have in your systems, and what a first forecast or early warning would need.
A first, rough estimate of what it could be worth, and what a pilot would look like.