🌐 Root Cause Analysis in the Industrial Domain using Knowledge Graphs: A Case Study on Power Transformers

📖 Introduction

This material, developed as part of a research presented in "Knowledge Graph-Driven Root Cause Analysis for Industrial Faults" (DOI: 10.1016/j.procs.2022.01.304), focuses on identifying, understanding, and correcting faults in industrial settings. The core component is an ontology that facilitates fault analysis through reasoning, classification, and advanced querying capabilities. 🧠

This work is based on the Arias & Mejia-Lara dataset: Dataset Link

🌍 Overview

In industrial environments, handling faults efficiently is critical due to associated costs. Traditional methods lack the capability to aid human operators in discerning fault causes and solutions. This ontology supports a knowledge graph-driven approach for root cause analysis, which includes:

1. Reasoning: Analyzing the current state of power transfromers. 🤖
2. Classification: Utilizing rules for failure classification. 🏷️
3. Querying: Leveraging graph-query languages for advanced data querying. 📊

📚 Ontology

The ontology onto-pw was applied in a power transformer case study, demonstrating its effectiveness in fault analysis.

Ontology URI

Base URI: http://test.org/onto-pw

Namespaces

• RDF: http://www.w3.org/1999/02/22-rdf-syntax-ns#
• XSD: http://www.w3.org/2001/XMLSchema#
• RDFS: http://www.w3.org/2000/01/rdf-schema#
• OWL: http://www.w3.org/2002/07/owl#
• SWRL: http://www.w3.org/2003/11/swrl#

Classes

1. Transformer: Represents the primary entity in the ontology.
2. Gas: A subclass of Transformer, representing various gases.
3. Property: General class for properties.
4. Estimation: Represents estimations.
5. Measurement: General class for measurements.
6. DBDS, Power_factor, Interfacial_V, Dielectric_rigidity, Water_content, Health_index, Life_expectation: Subclasses of Transformer, representing specific properties.
7. Hydrogen, Oxygen, Nitrogen, Methane, CO, CO2, Ethylene, Ethane, Acetylene: Subclasses of Gas, representing specific gas types.

Object Properties

1. is: Basic object property.
2. relatesWaterContentToMeasurement: Links Water Content to Measurement.
3. (Additional object properties are assumed to follow a similar pattern)

Datatype Properties

1. hasWaterLevel: Represents the water level, with domain as Water Content and range as integer.
2. (Additional object properties are assumed to follow a similar pattern)

Named Individuals

1. PW101 and PW102: Instances of Transformer with associated water levels.

AllDifferent Declaration

Ensures that specific classes like DBDS, Power_factor, etc., are recognized as distinct.

Usage

The ontology is structured to represent transformers and their properties, including various gas types and measurements. It allows linking specific transformer instances with their properties, such as water content. This ontology aids operators in the industrial domain, particularly in power transformer fault analysis, by providing a structured framework for reasoning, failure classification, and querying.

Additional Notes

The ontology is designed for extensibility and adaptability to various industrial scenarios. 🌟

📈 Classification

The file swrl-rules.py shows an example of how to perform classification.

Data Preparation and Analysis

• Libraries Used: pandas, owlready2, rdflib. 📚
• Data: The script initializes a dataset with various gases and other properties.

DataFrame Creation

• A DataFrame is created using pandas to structure the data. This DataFrame includes multiple columns for different gases and properties.

Ontology Creation

• An ontology is defined using the owlready2 library.

Ontology Population

• The script iterates through the DataFrame to create individuals in the ontology, assigning them properties based on the data.
• Each individual is named Individual followed by their index in the DataFrame.

Inference Rules and Reasoning

• Inference rules are established to classify individuals as Failure or NonFailure based on their LifeExpectation, HealthIndex, and PowerFactor.
• The Pellet reasoner is used for reasoning and to infer property values.

Output

• Finally, the script prints out lists of individuals classified as Failure and NonFailure.

🔍 Querying

The file query.py shows an example of how to perform querying.

Query Overview

• Purpose: To select transformers with water content levels above a specific threshold.
• Namespace Prefixes:
  • rdf: http://www.w3.org/1999/02/22-rdf-syntax-ns#
  • owl: http://www.w3.org/2002/07/owl#
  • xsd: http://www.w3.org/2001/XMLSchema#
  • Default (:): http://test.org/onto-pw#

Query Structure

• SELECT Clause: Retrieves ?transformer.
• WHERE Clause:
  • Matches any ?transformer that is of type :Transformer.
  • Associates ?transformer with ?waterContent using the :relatesToWaterContent predicate.
  • Extracts the water level ?level associated with ?waterContent using the :hasWaterLevel predicate.
  • Uses a FILTER to only include transformers where the water level ?level is greater than 100.

Query Goal

• The aim is to identify transformers within a given ontology that are potentially at risk due to high water content, as indicated by a water level exceeding 100.

📚 Reference

@inproceedings{martinez2021root,
  author       = {Jorge Martinez-Gil and
                  Georg Buchgeher and
                  David Gabauer and
                  Bernhard Freudenthaler and
                  Dominik Filipiak and
                  Anna Fensel},
  editor       = {Francesco Longo and
                  Michael Affenzeller and
                  Antonio Padovano},
  title        = {Root Cause Analysis in the Industrial Domain using Knowledge Graphs:
                  {A} Case Study on Power Transformers},
  booktitle    = {Proceedings of the 3rd International Conference on Industry 4.0 and
                  Smart Manufacturing {(ISM} 2022), Virtual Event / Upper Austria University
                  of Applied Sciences - Hagenberg Campus - Linz, Austria, 17-19 November
                  2021},
  series       = {Procedia Computer Science},
  volume       = {200},
  pages        = {944--953},
  publisher    = {Elsevier},
  year         = {2021},
  url          = {https://doi.org/10.1016/j.procs.2022.01.292},
  doi          = {10.1016/J.PROCS.2022.01.292}
}

📄 License

The material is provided under the MIT License.