moneo DataScience Toolbox – artificial intelligence
What is the SmartLimitWatcher?
The SmartLimitWatcher is the first tool of the moneo Data Science Toolbox, which offers solutions for production based on artificial intelligence. Users benefit from the permanent monitoring of the critical process value (target variable) with regard to the production quality or the plant condition (e.g. temperature, flow, vibration, current consumption). Anomalies in the target variable are detected automatically and at an early stage.
The SmartLimitWatcher is trained using historical data, which allows a permanent, reliable target/actual comparison between measured and predicted target value. The additional calculation of dynamic expectation ranges (confidence bands) for the target variable allows the permanent evaluation of the measured behaviour of the target variable as well as the automatic indication of deviations.
In contrast to static process value monitoring, with dynamic limit value monitoring the limit values depend on the current process state of the machine or system. Support variables describe the process state of the machine or system. Using a mathematical model, the dynamic limit values are calculated based on these support variables. In the event of a deviation (anomaly), a warning or alarm is automatically issued.
Difference between static and dynamic process monitoring
Prerequisites for successful use of the SmartLimitWatcher
- At least 2 process values are necessary:
- One process variable as the target variable to be monitored
- At least one other process value that is used as support variable
- The available data history should contain sufficient required operating states. (All cycles of a process should have been recorded several times. This is important for applications such as filters etc.)
- The process to be monitored must have a process relationship (non-linear/linear) between the target variable and the support variables. This relationship must be appropriately described by the available data. Due to this, all mechanically coupled systems are well suited. (The underlying process relationship does not have to be describable by formulae.)
- The process relationship that was taught for monitoring must also be valid in the future. (Example: Changes to the system to be monitored require new training.)
- No “artificial” support variables derived from the target variable, e.g. by calculation, may be used to monitor the target variable.
- Accordingly, there should not be too long a delay between the change in value of the target variable and those of the support variables. (Example: slow thermodynamic reactions)
- No processes with very “noisy” measured values should be considered, as in this case no precise model can be trained.
Application areas
The SmartLimitWatcher’s AI can be used in different ways for process monitoring. On the one hand for monitoring comparable machine components and on the other for monitoring individual add-on parts or measured variables.
1. Horizontal use
Monitoring based on comparable machine components.
Note on horizontal use
The connected machine components are integrated in a process or in the same plant and a physical dependency exists. An advantage is that you need but a few sensors or measured values to detect anomalies.
2. Vertical use
Detailed monitoring of a component using several measured values.
