A Fuzzy Rule-based Neuro-Symbolic Approach for Pipe...

A Fuzzy Rule-based Neuro-Symbolic Approach for Pipe... | AI Research

Key Takeaways

A Fuzzy Rule-based Neuro-Symbolic Approach for Pipe Severity Prediction in Sewer Networks introduces a framework to improve the transparency of automated sew...
Standard automated sewer pipe severity assessment relies on direct image classification, creating a "black box" where the link between visual defects and final severity scores remains implicit.
This study introduces a modular, fuzzy rule-based neuro-symbolic framework that bridges this gap by decoupling neural perception from symbolic reasoning.
The perception module utilizes a Swin Transformer to predict 14 multilabel inspection CODE degrees directly from images.
For reasoning, a DT, specifically Weka's J48, algorithm is trained on ground-truth CODEs and severity labels, and its paths are converted into 19 fixed IF--THEN rules.

Paper AbstractExpand

Standard automated sewer pipe severity assessment relies on direct image classification, creating a "black box" where the link between visual defects and final severity scores remains implicit. This study introduces a modular, fuzzy rule-based neuro-symbolic framework that bridges this gap by decoupling neural perception from symbolic reasoning. The perception module utilizes a Swin Transformer to predict 14 multilabel inspection CODE degrees directly from images. For reasoning, a DT, specifically Weka's J48, algorithm is trained on ground-truth CODEs and severity labels, and its paths are converted into 19 fixed IF--THEN rules. Inference operates via fuzzy logic: t-norm activations from CODE conditions are weighted by rule confidence and combined with corresponding s-norms to produce interpretable class evidence. We assessed Product, Łukasiewicz, and Hamacher operator pairs using a dataset of 3,244 images spanning five highly imbalanced severity classes. Ground-truth labels were robustly generated via consensus from five independent large language models analyzing original inspector notes. Our results show an improvement of accuracy, balanced accuracy, Macro F1 and MCC by 17.9%, 12.2%, 23.0%, and 17.3%, respectively, over image-only based classification. Overall, the framework combines competitive class-balanced performance with traceable reasoning from predicted CODE degrees to rule supports and severity evidence.

A Fuzzy Rule-based Neuro-Symbolic Approach for Pipe Severity Prediction in Sewer Networks introduces a framework to improve the transparency of automated sewer inspection. By separating the visual identification of pipe defects from the logical reasoning used to determine severity, the researchers aim to replace "black box" image classification with a system that provides traceable, human-understandable evidence for maintenance decisions.

Bridging Perception and Reasoning

The framework operates in two distinct stages. First, a neural perception module—based on a Swin Transformer—analyzes sewer images to predict 14 specific defect "CODEs." These codes serve as an intermediate semantic layer, representing the visual conditions of the pipe. Second, a symbolic reasoning module applies a fixed set of 19 IF–THEN rules to these predicted codes. These rules, derived from a Decision Tree (Weka’s J48), map specific combinations of defects to a final severity class. This design ensures that every severity prediction can be traced back to the specific defect codes and rules that triggered it.

Fuzzy Logic for Decision Support

Because visual evidence for a defect is often a matter of degree rather than a simple "yes" or "no," the framework uses fuzzy logic to process the neural output. The system calculates the truth value of each defect code and uses t-norm activations to combine these conditions within the rules. These results are weighted by the confidence of each rule and aggregated using s-norms to produce a final severity score. This approach allows the model to handle the ambiguity inherent in noisy, low-quality sewer imagery while maintaining a clear, logical path from image to decision.

Performance and Evaluation

The researchers evaluated the framework using a dataset of 3,244 images across five severity classes. To establish reliable ground truth, they used a consensus protocol involving five independent large language models to analyze original inspector notes. Compared to direct image-only classification, the neuro-symbolic approach showed improvements in accuracy (17.9%), balanced accuracy (12.2%), Macro F1 (23.0%), and MCC (17.3%).

Considerations for Implementation

The authors note that this framework is modular, meaning the neural backbone or the symbolic rule base can be adapted to different inspection standards without requiring a complete redesign. However, the system is not an end-to-end differentiable model; the symbolic rules are extracted once and remain fixed during inference. This separation is intentional, prioritizing the interpretability of the reasoning process over the joint optimization of the neural and symbolic components. The framework is specifically designed to address the challenges of high class imbalance in sewer data, where low-severity cases are significantly more common than critical ones.