Models and real time​

The development of our machine learning (ML) and analysis models usually takes place in the form of a prototype in Jupyter Notebooks. The advantage of Jupyter Notebooks is that they enable highly interactive and therefore agile development. The code is divided into different cells that can be executed independently of each other. This means that development can be carried out step by step. Jupyter Notebooks also offer the option of adding documentation to the code in the form of structured text, images and videos. In this way, you end up with not just the code for the prototype, but a comprehensive document. Below is an example of such a notebook.​

Once ML and other analysis models have been developed, the next step is to put them into practice. This often also means that the models have to function in real-time operation. The Analyzer4D software developed by QASS offers two options for integration at this point.​

  1. The integration of any Python code into the operator network.​
  2. The use of the analysis model detached from the Analyzer4D software.​

 Integration of any Python code in the operator network:  The operator network represents complex work sequences in the software that are started via various triggers or time-controlled. The operator network is made up of individual components, so-called operators, each of which represents a work step. The operators are connected to each other so that data can be exchanged. Finally, any Python code can be integrated via the so-called Python operator. Apart from the Python operators, the operator network is programmed graphically, which makes it much easier to use for people without programming knowledge.

However, analysis models can also be used separately from the Analzyer4D software. These are then installed in the form of additional software. Communication with the Analyzer4D software can then take place via a corresponding interface or completely independently by simply checking whether new measurement data is available.

Whether communication with the Analyzer4D software is required depends on whether the analysis model is also used for control.

To ensure the real-time capability of our models, it is important that the analysis itself takes less time than the actual processes being analysed. If a regulation is also linked to the analysis, the time requirements are even stricter. Here it is important that our analysis is able to deliver the results so quickly that a regulation can be implemented in good time. As a rule, we are talking about a few milliseconds. 

About Operator Network

Example of a real-time capable application​

An example of one of our real-time capable models is the analysis model that we use in our pipe pulling application. The task of the model is to analyse structure-borne noise data during the running process, but also to control the speed of the machine based on this data. This serves to avoid the formation of chatter marks on and in the pipe. To do this, we analyse various aspects of the measurement signal, such as the signal strength in the form of its amplitude, but also special signal characteristics. As the oscillation of the pipe, which ultimately leads to the formation of chatter marks, is already signalled in the signal image before damage occurs or is even audible to the human ear, we can reduce the speed of the machine immediately after detection and thus interrupt the oscillation process. This not only prevents damage to the pipe in the form of chatter marks, but also makes it possible to run at faster speeds on average. Furthermore, our real-time monitoring and control also means that the machine no longer needs to be continuously monitored by employees, as was previously the case.

More about tube drawing