TLDR: This research introduces an FBS (Function-Behavior-Structure) model-based method for accumulating maintenance records to improve failure-cause inference in manufacturing systems. It proposes a Diagnostic Knowledge Ontology that explicitly links deep knowledge (system structure and function) with shallow knowledge (failure events and their causal relationships). Experimental results on a LEGO car assembly line demonstrate that this approach significantly enhances the accuracy of failure-cause inference, especially in challenging cases with limited or varied maintenance data, by mitigating vocabulary mismatches and leveraging hierarchical system information.
In the fast-paced world of manufacturing, keeping production lines running smoothly is paramount. However, failures are an inevitable part of complex systems, and quickly identifying their root causes is crucial for maintaining efficiency and preventing costly disruptions. Traditionally, this task has relied heavily on experienced experts, but a growing shortage of such specialists means there’s an increasing need to empower non-experts with effective diagnostic tools.
Knowledge-based fault diagnosis methods aim to address this by building a knowledge base from expert insights and past failure analyses. However, existing approaches often face two significant challenges: effectively structuring both ‘deep knowledge’ (about the system’s design, structure, and function) and ‘shallow knowledge’ (empirical data on failures and their causal relationships), and ensuring these causal chains are long enough to trace back to the true root cause.
For instance, methods like Failure Mode and Effects Analysis (FMEA) are excellent for structuring deep knowledge about a system’s components and functions. Yet, they often fall short in providing sufficiently long causal chains of failures. On the other hand, maintenance records contain detailed, long causal chains of actual failures, but they are typically unstructured, written in free-form text by technicians, making them difficult to retrieve and reuse effectively.
A New Approach: Integrating Design and Maintenance Knowledge
A recent study proposes an innovative solution to these challenges: an FBS (Function-Behavior-Structure) model-based method for accumulating maintenance records. This approach, detailed in the paper “FBS Model-based Maintenance Record Accumulation for Failure-Cause Inference in Manufacturing Systems”, introduces a ‘Diagnostic Knowledge Ontology’ that bridges the gap between deep and shallow knowledge.
The core idea is to represent the manufacturing system using an FBS model, which defines how a system’s functions are realized by its behaviors, and how those behaviors are realized by its physical structures. This model is then integrated with a hierarchical understanding of functions, inspired by industry standards like AIAG & VDA FMEA methodology. The ontology organizes deep knowledge into five levels: Line Function, Process Function, Process Element Function, Behavior, and Structure, using a unified ‘has Part’ relationship to connect them hierarchically and by realization.
When a failure occurs, its details and causal relationships are recorded as ‘shallow knowledge’ instances and explicitly linked to the relevant parts of the FBS model. This creates a structured, machine-interpretable knowledge base that captures both the ‘what’ (system components and their roles) and the ‘how’ (the sequence of events leading to a failure).
How it Works in Practice
The process involves two main steps: First, at the system’s commissioning, an FBS model of the target manufacturing system is constructed. This model acts as a digital blueprint of the system. Second, during operation, every maintenance event and observed failure is documented and linked to the corresponding nodes within this FBS model. This ensures that each failure is not just a standalone event but is understood in the context of the system’s design and operation.
Also Read:
- Unlocking Complex Skills: How AI Bridges the Granularity Gap in Competency Modeling
- Proactive Safety: Predicting Robot Failures with Dual-Indicator Monitoring
Experimental Validation and Impact
The researchers tested their proposed method using a LEGO car assembly line, comparing its performance against a baseline method that used maintenance records without the FBS model integration. They used a Retrieval-Augmented Generation (RAG) inference engine to identify failure causes.
The results were particularly striking in ‘difficult cases’ – situations where only a few relevant maintenance records existed, and the wording used to describe the failure differed significantly from the input query. In these challenging scenarios, the proposed FBS model-based method significantly outperformed the baseline in both precision and recall. This improvement is attributed to the FBS model’s ability to handle variations in wording by incorporating system information and its explicit hierarchy, which helps narrow down the search space for relevant information.
This study highlights the immense value of reusing design-phase system definitions during the maintenance phase. By creating a structured, integrated knowledge base, manufacturing companies can not only shorten troubleshooting times but also establish a foundation for knowledge sharing across the entire engineering lifecycle, from design to maintenance and beyond.
