Operational reasoner

The Operation Reasoner (OR) computes assembly operations based on product characteristics and predefined rules. It requires context from sources like the AIToC Data Marketplace, housing rules defined in Answer Set Programming (ASP). These rules must cover all planning cases, simplifying with the Knowledge Editor. Model Feature Annotator identifies assembly-relevant features in 3D models for the OR. Using the state-of-the-art ASP solver, clingo, the OR generates operation sequences, storing them internally or externally in the Data Marketplace.

As mentioned above, the Operation Reasoner is implemented as an AJAN agent. In the AJAN Multi Agent System Framework, so-called SPARQL Behavior Trees are used for the internal behavior modeling of agents. They were used to implement the internal logic of the Operation Reasoner. This provides the Operation Reasoner with various REST-based interfaces to the outside world in order to obtain ASP, AutomationML and JSON-based rules and context information. As the data model used in AJAN is based on RDF, the data received is translated into RDF accordingly and stored in the Operation Reasoner agent's knowledge database. Once the data required for generating the operations sequences is available, the ASP Reasoner clingo used is executed to generate said sequences. The resulting plans are then stored externally via a REST API in the form of ASP files either in the Data Marketplace for future use or forwarded to other services for direct processing. The AJAN framework itself is implemented in JAVA using the RDFBeans and Springboot libraries. In addition to the backend implemented in JAVA, the AJAN framework also comes with a web editor based on JavaScript, NodeJS and EmberJS for agent modeling.

To be able to work with Operation Reasoner, the required AJAN framework can be obtained via GitHub (https://github.com/aantakli/AJAN-service, https://github.com/aantakli/AJAN-editor) or DockerHub (https://hub.docker.com/repository/docker/aantakli/ajan-service/, https://hub.docker.com/repository/docker/aantakli/ajan-editor/). AJAN can be executed under Windows, Linux and MacOS. The agent model for the Operation Reasoner is also available as an AJAN Package via GitHub (https://github.com/aantakli/AJAN-packages/blob/main/packages/Operation_Reasoner.ajan) and can be loaded, initialized and executed via the AJAN editor. After initializing the Operation Reasoner, ASP rules and context information can be transmitted natively, in the form of RDF, JSON or AutomationML via four different REST interfaces. The Operation Reasoner can also load data or ASP rules independently if the external storage location and credentials are transmitted to it via another JSON/REST interface. By using the RML library, this information is then translated into RDF and stored in the agent's database. The reasoning process can then be triggered via another RDF/JSON/REST interface. If an external storage location and credentials are specified, the generated operation sequences are stored in the specified storage location instead of in the agent database as mentioned above.

The Operation Reasoner is primarily intended for use by assembly planners. Due to its generic implementation, however, it is also suitable for domains outside of automatic planning, in which scheduling and automatic planning of action steps are required. This is because the planning itself is implemented using modeled ASP rules and these can be freely defined depending on the application.

The Java-based AJAN service is available under the LGPL 2.1 license and the JavaScript-based AJAN editor under the MIT license for free use as open-source software.