Neurosymbolic AI to Support Unstructured Data Processing in Process and Automation Engineering

Nicolai Schoch, Mario Hoernicke

Keywords: Process Plant Digital Twin, Process and Automation Engineering, Unstructured Data Processing, Ontology-enhanced Content Extraction

TL;DR: Making use of neurosymbolic AI capabilities for advancing the domain-knowledge-consistent automated processing of unstructured engineering data.

Abstract: Neurosymbolic AI can be an important enabler for AI-enhanced support of the unstructured data processing for process and automation engineering. Large subsymbolic models can shine with their semantic understanding of the contents of unstructured data sources, and symbolic representation can help to guide and guardrail the processing, so overall confidence in the generated outputs is reasonably high. With the Engineering Data Funnel, we present a neurosymbolic AI system, which uses large subsymbolic language and vision models to process unstructured engineering project data, while exploiting at the same time structured knowledge representation in engineering domain ontologies and thereon-based symbolic reasoning capabilities in order to provide reliable output for the deterministic further processing in traditional, established, mostly non-AI-based engineering tools.

Track: Main Track

Paper Type: Industry Abstract

Resubmission: No

Submission Number: 90