Predictive maintenance: making machine data usable
Without an environment that keeps machine data permanently and evaluable, every prediction model remains a prototype. Predictive maintenance rarely fails because of the model but because of the infrastructure beneath it: most projects fail not because of the analysis software but because the infrastructure is not designed for growing data volumes – cloud infrastructure for manufacturing and maintenance.
- BSI C5-audited & ISO 27001-certified – verifiable protection for machine and production data.
- Data centre in Germany – your production data remains within the German legal space.
- GPU and CPU scalable separately – add computing power for predictive analytics, keep continuous operation lean.
- Personal 24/7 service – support with connection, migration and operation.
What predictive maintenance requires technically
Predictive analytics is the method; predictive maintenance is its best-known application in manufacturing.
Time series, not tables
Machine data is time series data: many small values, high write rates, long history. A suitable database on ccloud³ handles write load and analysis at the same time.
Sovereignty over production data
Machine and production data reveal utilisation, processes and capacity. It stays within German jurisdiction, with no transfer to third countries.
GPU only when computing
Training needs GPUs, day-to-day operation does not. GPU and CPU capacity scale separately – you pay for the GPU only during training phases.
Older machines can connect too
Not every plant speaks MQTT. Named contacts with manufacturing experience advise on the path from the controller to the cloud.
Condition monitoring, predictive maintenance and predictive quality compared
Machine data arises where production takes place: modern plants deliver it via OPC UA or MQTT, older ones via retrofitted sensors and a gateway – the connection runs via a virtual private cloud, secured by a managed firewall.
Condition monitoring
Real-time condition monitoring: thresholds are monitored and reported. It describes what is happening now – without a forecast. The sensible intermediate step that provides the data basis for later prediction models.
Predictive maintenance
From the historical course it is derived when a component is expected to fail. Requires a robust data history – stored in a time series database on ccloud³, archived in S3 Object Storage.
Predictive quality
Predicts not the failure but product quality from process data. Combines machine, material and inspection data – the analysis tools for this on data analytics.
Industrial IoT (IIoT) as a platform
An IIoT platform is the industrial form of the Internet of Things and bundles connection, data storage, analysis and visualisation in one place. On centron, rented platforms run as well as self-operated stacks of message broker, time series database and visualisation – scaled via Kubernetes, operated on request by managed services; basics on IoT cloud.
Capture, store, analyse, predict
The path runs through four stages, and most projects fail at the second. Condition monitoring – observing actual values – is the sensible intermediate step: it produces the data basis on which a prediction model can be trained at all. centron supplies the right building block for each stage: ccloud³ for capture, S3 Object Storage for history and Cloud GPU for training.
- Capture – including from older controllers
- Store – time series with long history
- Analyse – condition monitoring as the intermediate step
- Predict – train models on GPUs
Predictive maintenance software and infrastructure: who is responsible for what
centron provides the infrastructure, not the analysis software. You bring the models and the user interface – as a vendor’s predictive maintenance software, as your own development or from an open-source toolkit. The link to planning is made via the ERP or maintenance module that knows bills of materials, stock levels and staff deployment – see ERP hosting. A time series database stores timestamped readings far more compactly than a classic relational database: a plant with a hundred measuring points at one-second intervals generates around 8.6 million records per day. Systems such as InfluxDB or TimescaleDB run on ccloud³ virtual machines or containerised via Kubernetes; block storage on the database VM holds the ongoing series, S3 Object Storage the history.
| Task | You or your software provider | centron |
|---|---|---|
| Sensors and data capture | selection, installation, pre-processing | — |
| Network connection and security | — | virtual private cloud, managed firewall |
| Data storage | data model and retention period | storage, backup, availability |
| Models | selection, training, evaluation | computing power including GPU |
| Operation | application logic and alerting | platform operation, monitoring, support |
The right centron products
Clinics and medical institutions typically implement IT security with these modules – combinable and extensible at any time.
- For time series databases
- Scale as needed
- German data centres
- €0.02 per GB per month
- Outbound traffic free
- Grows without pre-booking
- NVIDIA GPUs
- Scales separately from CPU
- Hourly billing available
What infrastructure does predictive maintenance require?
Predictive maintenance needs an environment that ingests machine data continuously, stores it permanently and keeps it analysable. The core building blocks are ccloud³ Virtual Machines from €3.11 per month for the time series database, complemented by S3 Object Storage for history and Cloud GPU from €92.59 per month for model training. The infrastructure runs in our own German data centres, certified to ISO 27001 based on IT-Grundschutz (BSI certificate BSI-IGZ-0773) and with an unqualified BSI C5:2020 Type 1 attestation. New accounts receive €200 in starting credit.
| Module | Price |
|---|---|
| ccloud³ | from €3.11 / month |
| S3 Object Storage | from €5.00 / month |
| Cloud GPU | from €92.59 / month |
Frequently asked questions
What is predictive maintenance?
What is predictive analytics and how is it related?
What data does predictive maintenance need?
What belongs on the machine and what in the cloud?
Why a time series database rather than a classic one?
Can older machines be connected?
Does production data stay in Germany?
Do I need GPUs for predictive maintenance?
How do I start without investing in hardware first?
More for manufacturing
Get expert advice
Tell us about your project. Our sales team will get back to you within one business day.
The consultation is free of charge and without obligation.