Support

Frequently asked questions

Predictive Maintenance

What is predictive maintenance?

Predictive maintenance is a maintenance strategy in which machine data is continuously analyzed to detect impending faults early and predict failures. It is about knowing which fault will occur and when. This allows for maintenance to be performed when it is actually needed: not according to a rigid schedule and not only after damage has occurred.

How does predictive maintenance work?

Sensors and controllers continuously provide data such as temperature, vibration, pressure, or power consumption. An AI learns the machine's normal operating state from this, detects deviations, and reports them early so that maintenance can intervene in a targeted manner.

What does machine maintenance involve?

According to DIN 31051, maintenance comprises four measures: inspection, servicing, repair, and improvement. Predictive maintenance primarily supports inspection and maintenance planning by continuously evaluating machine condition.

What are the benefits of predictive maintenance?

Fewer unplanned downtimes, lower maintenance costs, longer component service life, better planning for operations and spare parts, and a noticeable reduction in the workload for the maintenance team.

What is the difference between predictive, preventive, and reactive maintenance?

Reactive maintenance repairs only after a failure has occurred. Preventive maintenance performs servicing at fixed intervals, regardless of the actual condition. Predictive maintenance performs servicing based on condition data and forecasts—as early as necessary, as late as possible.

What is the difference between condition monitoring and predictive maintenance?

Condition monitoring tracks the current state of a machine in real time. Predictive maintenance goes a step further, using this data and AI to forecast future faults and maintenance requirements.

Machine Monitoring

What is automated machine monitoring?

Automated machine monitoring continuously captures and evaluates machine data such as temperature, vibration, pressure, or current consumption, eliminating the need for manual inspections or manual readings. Anomalies are automatically detected and reported. This gives the maintenance team 24/7 transparency regarding the machine's condition and significantly reduces their workload.

How does AI-supported machine monitoring work?

The AI learns the normal behavior of each machine from ongoing sensor and operational data. It compares new data in real time against this learned behavior and summarizes deviations in an anomaly index (0–100). If abnormal behavior persists for the defined tolerance period (slightly abnormal, abnormal, or highly abnormal), you will automatically receive a notification with an AI explanation.

What does anomaly detection mean for machines?

Anomaly detection identifies failure patterns in machine data that deviate from normal behavior – such as unusual vibrations, temperature fluctuations, or current consumption trends. Such anomalies are often early signs of wear, damage, or process disruptions, long before a failure occurs.

How does AI detect machine faults?

The AI evaluates all of a machine's measured variables collectively and detects when individual values or their interaction deviate from the learned normal behavior. Dynamic limits instead of fixed thresholds distinguish normal operational changes from critical developments. For every anomaly, the dashboard displays the affected sensors and a brief description.

Suitability / Use Cases

Which machines is the aiomatic maintenance software suitable for?

For all machines with motors or rotating components that provide sufficient data via sensors – e.g., pumps, compressors, agitators, fans and blowers, mills, extruders, and conveyor belts.

Which target groups is the maintenance software aimed at?

For companies where unplanned downtime causes high costs or productivity losses, especially manufacturing companies with complex, data-rich equipment in continuous operation.

When is aiomatic software the right choice for your maintenance strategy?

When you are looking for a solution that continuously analyzes machine data and detects anomalies in real time. The software does not handle maintenance schedules, spare parts ordering, or parameter optimization.

What are the use cases for predictive maintenance?

For example, monitoring geared motors, pumps, compressors, conveyor belts, agitators, continuous furnaces, or continuous printing and coating systems. This allows for the early detection of issues like bearing or gearbox damage.

How can predictive maintenance be used for motors, pumps, and gearboxes?

For motors, performance or current consumption, speed, and vibration are primarily analyzed; for pumps, pressure, flow rate, and temperature are also monitored; for gearboxes, the focus is on vibration and temperature. The AI detects deviations from normal behavior, such as early-stage bearing or gearbox damage.

How much data expertise is required on the company side?

You don't need your own data scientists – aiomatic provides a ready-to-use AI solution. However, it is important to have your process and machine experts as well as the maintenance team involved to provide context.

Does machine data need to be recorded digitally already?

Yes, machines should have a medium to high level of digitalization, ideally with data collection at fixed intervals. If this is missing, sensors can be retrofitted using hardware from our partners KSB or Avibia.

Which industries benefit most from predictive maintenance?

Primarily industries with continuously operating, critical equipment – e.g., energy, pharmaceuticals, chemicals, food, automotive, mechanical engineering, water management, as well as logistics and ports.

Questions about aiomatic's predictive maintenance software

What types of data does aiomatic's software analyze to support predictive maintenance?

Sensor data (e.g., temperature, vibration, pressure, speed, current consumption) as well as operational and process data (e.g., runtimes, operating modes).

In which data formats can the software process data?

The software processes structured time series with timestamps: numerical measurements, binary signals (e.g., on/off), and categorical values (e.g., operating mode, product variant). Free text, images, video, or audio are not yet part of the analysis.

Where is aiomatic's predictive maintenance software used?

In production environments with continuously running machines and systems that regularly provide sensor or process data. This ranges from, for example, pumps, motors, and compressors to conveyor belts or continuous furnaces. Especially in industries such as energy, pharmaceuticals, chemicals, food, automotive, mechanical engineering, water management, as well as logistics and ports.

Where exactly is the data stored and analyzed?

Data is transmitted in encrypted form and stored and analyzed in an ISO/IEC 27001 certified cloud environment.

Is aiomatic GDPR and EU AI Act compliant? What certifications are in place (e.g., ISO 27001)?

Data processing is carried out in compliance with data protection regulations. The solution is hosted by a cloud provider whose environment is ISO/IEC 27001 certified. You can find further details on our security page.

How does the AI distinguish between an anomaly and a normal operational change, or between normal wear and a critical change in condition?

In practice, we do not distinguish between normal wear and tear and a critical state. The transition is fluid and becomes apparent through a steadily declining anomaly index

What different solutions does aiomatic offer?

A one-day workshop, historical data analysis, a proof of concept, and a SaaS model for continuous operation – with the packages Predict (AI software), Retrofit (sensor retrofitting for rotating machinery), and Retrofit & Predict (a combination).

Which use cases does it specifically cover: condition monitoring, anomaly detection, diagnosis, failure prognosis, or maintenance optimization?

The focus is on condition monitoring, anomaly detection, and early fault detection, including AI explainability. Once a threshold and tolerance time are set, critical patterns can be identified well in advance.

How the AI works: Detection, diagnosis & prognosis

What specific types of failures can the software detect?

Typical examples include mechanical failures in rotating machinery such as bearing, gearbox, and motor damage, imbalance, or wear, as well as process deviations, such as unusual temperature, pressure, or current profiles. The prerequisite is that the failure is reflected in the available data. High-frequency vibration data is usually required for bearing and gearbox damage.

Does the system only detect deviations, or does it also identify probable causes of failures?

Both: The software detects deviations and provides an AI explanation for every anomaly with a brief description of which sensors are showing irregularities. The final assessment is made by your specialist staff.

Which failure patterns are already pre-trained, and which must be trained on a customer-specific basis? Are there pre-configured models?

The AI does not require pre-trained failure patterns or labeled data: it learns the normal state of each machine and detects deviations from it.

Can custom failure patterns, rules, and empirical values be added?

Yes. In the logbook, you can document maintenance, malfunctions, product changes, or actual damage as events. We can add customer-specific event types if needed. You can also provide feedback for every alert, e.g., "impending defect," "machine setting," or "not an issue." Events and feedback flow directly into the training of the models. Sensitivity, thresholds, and tolerance times are individually adjustable.

How or by whom are the models created?

The aiomatic team creates and trains the models based on your machine and process data. We determine the relevant influencing factors ourselves. Input from your experts is welcome but not a requirement.

Are models trained per machine, per machine type, or across the board?

A model learns the normal behavior of a sensor. Taken together, the anomalies detected by a machine’s models indicate the damage.

Does the solution work without sufficient historical failure data?

Yes. Historical failure data is not required; the AI learns the machine's normal state during live operation. Data collection begins on day one, and reliable models are typically available after about three months. Initial results are available even sooner.

How does the solution account for different operating states such as downtime, partial load, full load, and product changes?

The software processes categorical context data such as operating modes, recipes, or product variants, thereby recognizing which behavior is normal in each respective state.

Can multiple measured variables be evaluated and correlated together?

Yes. The AI evaluates all channels of a machine collectively (typically 5–200 data channels) and also detects deviations in the interaction of multiple measured variables.

How are anomaly limits defined, adjusted, and documented?

Instead of fixed limits, the AI uses dynamic thresholds. The threshold and tolerance time for anomalies can be adjusted individually.

Can the user understand which data and influencing factors led to an alert?

Yes. For every notification, the dashboard displays an anomaly summary (AI explanation) showing the affected sensors.
This makes it easy to understand which data led to the notification.

How is a forecast validated and improved during ongoing operation?

Through continuous learning from new data and your feedback on notifications. This allows the models to become steadily more precise during ongoing operation.

What happens to the model after modifications, repairs, or process changes?

Ideally, the model simply continues to run, as the machine has been returned to its baseline state. If that is not the case, the new normal state is learned or the model is retrained. Please inform us of such changes so they can be correctly classified.

Historical data analysis

What data is required for the analysis?

For the analysis, we need:
‍
Operational data: Information on operating conditions, such as running times, system load, production rates and environmental effects.
‍Sensor data: Physical measurements such as temperature, oscillations, speed, or power consumption.‍
‍Maintenance and fault data: Historical records of failures, maintenance measures and their timestamps.
Technical specifications: Information on machine manufacturers, models and performance data.

In which format should the data be provided?

We prefer consistent file formats, in particular:
‍
CSV files: Time series data with clear column headings.
‍Alternatively: Excel or database exports (such as SQL, TimescaleDB). It is important that all data is provided with synchronized time stamps.

Which information is particularly important when it comes to fault and maintenance data?

The fault and maintenance data should include the following details:
‍
Timestamp: Timestamp for the start and end of the fault or maintenance
Description of the cause: Description of the cause (e.g. mechanical defect, sensor problem)‍
Downtime details: Downtime information (planned vs. unplanned outages)
Measures: Remedial measures and their duration

What additional information helps with data analysis?

‍Machine specifications: Manufacturer, model, year of manufacture, tolerances and performance ranges
Sensor information: Position, type, manufacturer, and potential redundancies‍
Operating time pattern: Continuous or batch production, planned maintenance cycles. This information helps to correctly interpret the data and optimize analysis results

PoC

When is a proof of concept (PoC) useful?

If you are still unsure whether our solution is the right fit for your requirements or if your data basis is sufficient, we offer a proof of concept (PoC). This allows you to test our software in your specific environment and evaluate the data basis before committing to comprehensive live monitoring via our SaaS solution.

What is the process for a proof of concept (PoC)?

In month 1, we set up the data infrastructure and establish all connections. Month 2 is dedicated to visualization and data collection, while in month 3, we develop and train the AI models. From month 4 onwards, the system goes live, with continuous optimizations to ensure the best results.

What results does a proof of concept (PoC) provide?

The focus of the PoC is on establishing a solid data basis. Efficiency gains and cost reductions follow in later phases once the models are further optimized. We recommend defining specific, realistic goals for the PoC.

How does aiomatic help in selecting economically viable pilot machines?

In our workshop and the free machine and use-case analysis, we work together to evaluate the data basis, criticality, and downtime costs of your equipment to select the pilot machines with the greatest economic potential.

Project workflow

How does the introduction of predictive maintenance begin in a company?

Start with an as-is analysis: identify relevant machines and existing sensor data. We help you identify the use case and guide you step by step through the successful introduction of predictive maintenance.

What are the challenges in implementing predictive maintenance?

Challenges include integrating various data sources, selecting and potentially retrofitting suitable sensors, the need for qualified personnel for data analysis, and the initial investment costs. The expert team at aiomatic supports you through every step.

What does a typical predictive maintenance project look like at aiomatic, and how long does it take to get the first analysis results?

Month 1: Setting up data infrastructure and connectivity; Month 2: Visualization and data collection; Month 3: Development and training of AI models; From Month 4: Live operation with continuous optimization. You will see initial data in the dashboard as soon as the connection is established, and reliable AI models are typically available after about three months.

What personnel resources or effort are required on the company side?

You will need at least one person responsible for maintenance, an IT contact for network permissions, and, if necessary, someone from production for OPC UA information. In one of our retrofit projects, 20 use cases were implemented with only about 6 hours of effort from the customer.

How reliable are the alerts – and how does aiomatic prevent false alarms?

The AI uses dynamic limits instead of fixed thresholds and takes operational context, such as recipes, into account. Sensitivity and tolerance times are adjustable, and your feedback on alerts continuously improves the models.

How are notifications handled when machine behavior anomalies occur?

Automatically via email as soon as a machine behaves in a significantly unexpected way. Additionally, you can view the anomaly overview with AI-generated explanations in the dashboard.

What kind of support do I receive after go-live?

After go-live, we continuously optimize the models and remain available as your point of contact. Regular meetings with your users are included in the price. We coordinate more frequently during the initial phase, followed by scheduled weekly meetings. During these, we discuss anomalies, assist in evaluating alerts, and answer your questions.

Integration and Interfaces

Can aiomatic be integrated into existing systems?

Yes. The standard is connection via OPC UA (using aioConnect) or MQTT with structured JSON messages. Depending on the environment, REST or Modbus TCP are also possible. Data can be read from PLCs, SCADA, gateways, or IoT platforms.

Can existing sensors be used?

Yes. Existing sensors are used as long as their data is available digitally. Retrofitting is only necessary if suitable sensor technology is missing.

In what form/format must the data be provided?

With aioConnect, the standard is OPC UA, or MQTT as structured JSON messages – each with a timestamp and measured value. For historical analysis, CSV files or Excel/database exports with synchronized timestamps are suitable.

What happens if the data quality is insufficient?

Before the project starts, we generate an automated data quality report. It evaluates each channel individually, checking for issues such as constant values despite dynamic processes, data gaps, or insufficient resolution. Where the data is insufficient, the report explains the reasons and provides specific recommendations. In addition, you can easily retrofit a vibration sensor at any time using KSB Guard.

How much historical data and what data quality are required to get started?

Historical data is not a prerequisite: Data collection begins from day one of the connection, and reliable models are typically available after about three months. It is important that the data is recorded regularly and can be clearly assigned.

Sensor retrofit

What does sensor retrofitting mean?

Sensor retrofitting means equipping existing machines with sensors after the fact so that their condition can be digitally recorded and analyzed. Together with our hardware partners, we can easily retrofit sensor technology.

Can older machines be retrofitted?

Yes. Even older machines without digital interfaces can be retrofitted with sensors and connected, for example via an edge device with an LTE connection, independent of your IT. For instance, we offer sensor retrofitting in collaboration with our hardware partners KSB or Avibia.

Which measurement parameters are relevant for rotating machinery?

The two most important parameters for rotating machinery such as pumps, motors, fans, or gearboxes are vibration and temperature. With our retrofit package, we can easily add the appropriate sensor technology.

Can I install the sensors myself, or do I need specialized personnel?

The plug-and-play sensors from our partner KSB are designed for easy installation and you can mount them yourself. Of course, we are happy to provide support if needed.

Workshop

How much time should I plan for a workshop with aiomatic?

Plan a day – our team will be happy to visit you personally at your location. We analyze your data basis, develop a strategy and create a project plan for implementation.

Who should attend the predictive maintenance workshop?

Maintenance, IT and production specialists as well as decision-makers – ideally everyone involved in implementation and use.

Does the workshop commit me to buying the software?

No, the workshop is of course non-binding. It is used to analyze your requirements and develop a suitable concept.

What's next after the workshop?

If you decide to work together after the workshop, we start with a proof of concept (PoC) to test the solution in your environment and validate the added value of predictive maintenance software.

Pricing and effectiveness

How much does predictive maintenance with aiomatic cost (per sensor/channel)?

The price depends on the selected package and the number of monitored data channels.

Can additional machines or locations be added later?

Yes. The SaaS model is scalable. Additional sensors, machines, or locations can be added easily at any time.

Is predictive maintenance also useful for small or non-critical machines, or is it only worth it for critical equipment?

The greatest benefit is seen in machines whose failure causes high costs or production losses. It is also worth taking a closer look at non-critical individual machines.

Is predictive maintenance worth it if there are currently hardly any failures?

Yes. Even rare failures can be very expensive. In addition, predictive maintenance helps to optimize maintenance intervals, relieve the burden on personnel, and use spare parts in a targeted manner. Our ROI calculator shows the benefits for your equipment.

How can predictive maintenance software help reduce costs?

By predicting failures, avoiding unplanned downtime, extending service life through targeted maintenance, optimizing maintenance cycles, and increasing equipment availability.

What does an unplanned machine breakdown cost and how much can downtime be reduced through predictive maintenance?

This depends heavily on the facility and the industry. In one customer example, a single component failure cost around €14,300. The extent to which downtime can be reduced depends on the use case. You can calculate your own downtime costs using our ROI calculator.

Further questions

Which predictive maintenance providers are there in Germany?

In Germany, there are specialized AI providers like aiomatic from Hamburg, providers of traditional condition monitoring, and industrial corporations with their own IoT platforms. They differ primarily in terms of hardware dependency, setup effort, and their focus on specific machine types.

Which predictive maintenance software is the best, and which is suitable for small and medium-sized enterprises?

The best solution depends on your machines, data, and resources. For small and medium-sized enterprises, the key factors are a quick implementation without a major IT project, the use of existing sensors, minimal internal effort, and a scalable SaaS model.

Which predictive maintenance solution is the best for pumps?

Solutions that specifically evaluate temperature and vibration are suitable. Together with KSB, aiomatic offers an uncomplicated plug-and-play solution, including the appropriate sensor technology from the pump manufacturer KSB.

What does Maintenance 4.0 mean?

Maintenance 4.0 refers to digital, data-driven maintenance in the context of Industry 4.0. Connected machines continuously provide status data, which is analyzed using AI to plan maintenance in a predictive and needs-based manner.

How is AI used in maintenance?

AI is primarily used for automatic anomaly detection. It learns normal machine behavior from sensor data and reports deviations early on, allowing teams to intervene in a targeted manner.