Automated Reasoning for City Infrastructure Maintenance Decision Support

Lijun Wei, Derek R. Magee, Vania Dimitrova, Barry Clarke, Heshan Du, Quratul-ain Mahesar, Kareem Al Ammari, Anthony G. CohnIJCAI 2018 · DOI

TL;DR

This paper introduces ATU-DSS, an interactive decision support system that uses automated reasoning to help manage city infrastructure. By integrating real-time data, expert knowledge, and an inference engine, it predicts potential consequences of infrastructure issues, handles uncertainty, and suggests further investigations, aiming to improve maintenance decisions.

What Problem Does This Paper Address?

Maintaining urban infrastructure is a complex and critical task. As infrastructure ages and urbanisation increases, the risk of disasters like sinkholes and service disruptions grows. This complexity stems from the need to integrate vast amounts of diverse, often incomplete, and uncertain data from multiple sectors (e.g., roads, utilities, ground conditions, weather, traffic). Decision-makers, such as incident managers and engineers, struggle to collect and interpret this information to make informed maintenance choices. Existing approaches often lack the ability to proactively predict consequences, handle uncertainty effectively, or provide clear explanations for their reasoning.

Methods

The ATU-DSS system employs a three-component architecture:

  1. Knowledge Model: This component captures multi-sector expertise in a computer-processable format. It includes:
  2. City Infrastructure Ontologies: Defining concepts for roads, buried assets, and ground conditions.
  3. Trigger Ontology: Specifying events that necessitate infrastructure management decisions (e.g., road cracks, pipe leakage).
  4. Investigation Ontology: Encoding geophysical techniques for asset property measurement. These ontologies provide a common vocabulary for data integration.

  5. Integrated Database: Stores real-world data, including historic weather data (British Met Office) and ground condition datasets (British Geological Survey).

  6. Inference Engine: This is the core reasoning component. It links the knowledge base and real-world data to infer potential consequences. Key features include:
  7. Predictive Rules: Defined by domain experts using a confidence vector system (e.g., Impossible, Very Unlikely, Likely, Definite) to handle qualitative uncertainty.
  8. Automated Diagnosis Rules: Generated to infer possible causes from observed effects.
  9. Handling Incomplete Data: The system can fire rules even if some premises are missing by assuming worst-case scenarios.
  10. Reasoning Chain and Explanation: The system can explain its conclusions by showing the reasoning chain and potential consequences in an impact matrix (likelihood vs. severity).
  11. Interactive Updates: Users can add or update missing data, triggering an automatic re-activation of the reasoning process.

The system's workflow begins with a reported trigger, followed by real-time data retrieval, feeding into the inference engine for automated reasoning.

Datasets

The paper mentions the use of:

  • Historic weather data sourced from the British Met Office.
  • Ground condition datasets sourced from the British Geological Survey.
  • Real-world data of infrastructure assets and contextual information from various data owners.

The paper does not introduce new benchmark datasets but rather integrates existing real-world data.

Key Insights and Ideas

  • Bridging the Gap Between Data and Decisions: The system effectively bridges the gap between raw, disparate data and actionable maintenance decisions by providing a structured reasoning process.
  • Handling Qualitative Uncertainty: The use of qualitative confidence levels (e.g., "Likely," "Very Unlikely") is a practical approach to representing expert knowledge when precise probabilities are difficult to obtain.
  • Proactive Risk Assessment: By predicting potential consequences, the system shifts the focus from reactive problem-solving to proactive risk management and maintenance planning.
  • Transparency in Reasoning: Providing reasoning chains and explanations is crucial for building trust and enabling users to understand why a certain consequence is predicted, facilitating better decision-making.
  • Worst-Case Scenario Assumption: The ability to make assumptions about missing data to explore worst-case scenarios is a powerful feature for identifying critical risks that might otherwise be overlooked.

Key Results

The paper demonstrates the ATU-DSS through multiple real-world scenarios, showcasing its ability to:

  • Predict potential consequences and risks based on site-specific data and expert knowledge.
  • Provide explanations for each predicted consequence.
  • Cope with incomplete and uncertain data by making assumptions.
  • Suggest further investigations.
  • Show how modifying assumptions (parameter values) can lead to different consequences.

A specific example is given of a "leaking water pipe" scenario, illustrating how it can lead to consequences like "TrenchBackfillWaterContent increases" and potentially "SuddenRoadCollapse," with associated likelihood and severity levels presented in an impact matrix.

Limitations and Future Work

Limitations: * The paper notes that experts found it difficult to formulate rules with numeric probabilities, leading to the adoption of qualitative confidences. This might limit the precision of some predictions. * The system's effectiveness relies heavily on the quality and completeness of the input data and the accuracy of the expert-defined rules.

Future Work: * The paper suggests that a video demonstrates the complete workflow, implying that further development and showcasing of the system are ongoing. * The authors aim to improve the system's ability to handle complex decision-making requiring multi-sector knowledge, suggesting continued expansion of its scope and capabilities.

Summary updated

BibTeX
@inproceedings{wei2018automated,
  author = {Lijun Wei and Derek R. Magee and Vania Dimitrova and Barry Clarke and Heshan Du and Quratul-ain Mahesar and Kareem Al Ammari and Anthony G. Cohn},
  title = {{Automated Reasoning for City Infrastructure Maintenance Decision Support}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2018},
  doi = {10.24963/ijcai.2018/868},
}
Figure