Automating SAPPhIRE Modeling for Product Systems and Extending SAPPhIRE Model to Service Representation

Abstract

Creativity in product design sparks innovation, leading to unique solutions that meet evolving consumer needs. This innovation drives economic growth by creating new markets and job opportunities. Moreover, creative product design can transform society by enhancing quality of life, promoting sustainability, and addressing social challenges. Design by Analogy is a proven method to unlock creativity in product design. Researchers have developed support for Design by Analogy in Product Design using ontologies, like SBF and SAPPhIRE, that describe system functioning. These ontologies provide a powerful abstraction of natural or technical systems, aiding in creative design ideation. Due to their significant role in creative design ideation, databases of ontology-based models for biological and technical systems were developed. However, creating structured database entries through system models using an ontology requires the time and effort of experts by sourcing knowledge about the system working from multiple technical documents. Researchers worked towards developing methods that can automatically generate representations of systems from documents using ontologies by leveraging machine learning techniques. However, these methods use limited, hand-annotated data for building machine learning models and have manual touchpoints that are not documented. While opportunities exist to improve the accuracy of these models, it is more important to first understand the complete process of generating structured representations using an ontology first and then automating it. This research employs the SAPPhIRE model, a comprehensive causal framework for representing both technical and natural systems, which has proven effective in facilitating creative ideation during the conceptual design phase. The study introduces a novel method, along with a set of rules, for extracting information relevant to the constructs of the SAPPhIRE model from natural language descriptions of technical systems. The proposed method interprets information in the context of the entire description by first identifying system interactions involving material, energy, and information. It then constructs a causal representation of each interaction using the SAPPhIRE ontology. Developed through an iterative process, the method was refined and validated through successive user trials. These trials, conducted with both expert and novice users of the SAPPhIRE model, demonstrated that the method enables accurate and consistent extraction of relevant causal information from natural language inputs. Building on the previously proposed method, this research presents its automation through the development of a computational method for extracting information relevant to the SAPPhIRE model of causality. The system integrates dependency parsing of natural language processing techniques along with a lexical database and knowledge graphs to identify and store key words and phrases from technical descriptions. It then applies rule-based reasoning to map this information to the constructs of the SAPPhIRE model. Unlike supervised learning approaches, this method does not require large, annotated datasets for training. The accuracy and effectiveness of the automated system were validated through expert evaluations conducted by SAPPhIRE specialists. Developing a SAPPhIRE model for a technical or natural system typically requires synthesizing knowledge from multiple technical documents to understand system behavior. To address this challenge, this research explores the use of Large Language Models (LLMs) for generating technical content aligned with the constructs of the SAPPhIRE model of causality. A key concern in this context is the issue of hallucination—where LLMs generate plausible but inaccurate information. To mitigate this, the study investigates hallucination suppression through Retrieval-Augmented Generation (RAG), wherein the LLM is provided with curated reference material to ground its outputs in verified scientific knowledge. The findings indicate that both the content and structural organization of the reference knowledge significantly influence the accuracy and relevance of the generated content with respect to SAPPhIRE constructs. Driven by intense competition and the influence of Industry 4.0, manufacturers are increasingly adopting servitization, shifting from offering standalone technical products to delivering integrated product-service systems (PSS). Concurrently, the emergence of cyber-physical systems is enabling products and services to become 'smart' by leveraging operational data to enhance value delivery. While the SAPPhIRE model of causality has been effectively applied to represent the functioning of technical systems and support various design activities, such as analysis, synthesis, and novelty assessment, its application to services and integrated product-service offerings remains underexplored. To address this gap, this research investigates how the SAPPhIRE model can be extended to represent services by capturing the causal relationships among service components. The study demonstrates how this extended model can be used to analyze existing service configurations and support the synthesis of new ones. Furthermore, the SAPPhIRE-based representation of services is compared against an established system modeling method previously used for service modeling.