Developing prototypes for an automated control programme in PLCopen from state machines and a learning-capable Part Average Analysis for a real production environment.
The project consists of two independent parts
-System configuration PLCopen
-Average Learning (PAL)
In both parts the goal is to develop a prototype to prove the feasibility of the approach.
System configuration PLCopen
Investigating possibilities for user-driven and partly automated configuration of elements and glue logic. To validate the results of the analysis a prototype for a PLCopen XML Export was implemented as proof-of-concept. The integration requirements for implementation of the prototype into the engineering software used by the project partner were investigated and defined. The prototype was tested based on an actual example of automation of a bio-waste treatment plant. With a software modelling tool the plant was modelled in the form of state machines. This model was implemented in PLCopen and can therefore be used for all applications that are compliant with AutomationML/PLCopen. The possible applications of the PLCopen export were then tested for common SPS systems (CoDeSys, Beckhoff TwinCAT, Siemens TIA Portal).
The prototype of a Part Average Analysis (PAA) capable of learning was developed. The aim of a PAA is to observe anomalies in functions or individual parameters to reliably detect hidden flaws during the production process. As a learning algorithm a neural network was chosen since it allows the training on real-time data and thus a learning effect during ongoing operation. The algorithm was trained on generated data of a production environment and then validated with real time data to identify conspicuous data. The neural network showed an error detection of at least 99% with the test data.