Urban mining, understood as the process of recovering materials and components from buildings at the end of their life cycle, increasingly relies on intelligent systems for pre-demolition assessment. These systems must not only predict accurately but also generate defensible decisions for regulators and auditors. The combination of knowledge graphs (KG) and explainable AI (XAI) offers a promising path to achieve that balance.
Urban mining has become a pillar of the circular economy, enabling the recovery of steel, copper, wood, and other materials from obsolete buildings. However, the pre-demolition assessment process is complex: it involves regulations, structural safety, environmental impact, and material valuation. Auditors need tools that not only provide predictions but also explain how each conclusion was reached.
Artificial intelligence (AI) can analyze large volumes of data from plans, inspections, and sensors, but black-box models generate distrust. On the other hand, knowledge graphs structure building information into entities and relationships – such as construction components, their properties, and applicable regulations – but lack AI's predictive capability. The complementarity between both technologies enables systems that offer both accuracy and transparency.
In this article, we explore how the integration of KG and XAI can transform urban mining, and how companies like Q2BSTUDIO are developing custom software applications for this sector. Additionally, we will mention complementary services such as cloud, cybersecurity, and BI that enhance these solutions.
A knowledge graph represents the construction domain: materials, door types, fire protection systems, local regulations, etc. Each node is an entity and each edge a relationship. For example, a fire door has properties like fire resistance, location, and installation date. This graph allows auditors to query the building's history and verify regulatory compliance.
Explainable AI, meanwhile, not only generates a classification (e.g., 'door suitable for reuse') but also provides reasons: feature weights, counterexamples, or underlying rules. Techniques like LIME, SHAP, or counterfactuals help understand why the model made a decision. However, these explanations can be inconsistent if not anchored in a rich semantic context.
This is where KGs act as a 'semantic skeleton'. By integrating AI explanations with graph relationships, contextualized justifications are obtained. For example, an explanation may indicate that the determining factor was 'year of manufacture', and the KG links that attribute to the regulation in effect that year, allowing the auditor to evaluate the decision's validity.
Integration modes are varied. One consists of 'lifting' model predictions to graph concepts, assigning semantic labels. Another is 'constraining' the model's search space using graph restrictions, avoiding impossible predictions. Also, 'typing' explanations according to graph categories, and 'revising' results through graph queries. Although these terms come from literature, their practical application depends on the specific implementation.
In practice, an urban mining system based on KG and XAI could work as follows: an auditor uploads images of a fire door; a computer vision model classifies its condition; the system extracts from the KG the expected features per regulation; then, an XAI module generates an explanation showing which image areas influenced the decision; finally, the auditor can contrast that explanation with graph data (installation date, certifications) to issue a defensible report.
This flow requires robust infrastructure: cloud storage (AWS or Azure) for sensor and plan data, real-time AI capabilities, and cybersecurity protocols to protect sensitive building information. Additionally, auditable report generation can be supported by Business Intelligence tools like Power BI, which visualize explanations and graph data in a comprehensible way.
Q2BSTUDIO, as a software and technology development company, offers precisely this: artificial intelligence solutions integrated with knowledge graphs, tailored to each client's needs. Its services include creating AI agents that interact with the KG to answer audit questions, automating evaluation processes, and consulting on cloud and cybersecurity.
A concrete case is the evaluation of fire doors in a commercial building. The knowledge graph contains specifications for each door (model, year, certifications). An image classification model determines if the door meets current standards. The explainable AI indicates that the decision was based on the presence of fireproof sealing and frame condition. The auditor, through a web application, can navigate the KG to verify the last inspection date and applicable regulations, thus generating a complete and defensible report.
This approach not only improves evaluation accuracy but also reduces legal risk and increases regulator confidence. The key lies in complementarity: the KG provides context; XAI provides transparency; together they enable auditors to make informed and justifiable decisions.
The adoption of these technologies in urban mining is still in early stages, but the potential is enormous. Companies that invest in integrated KG and XAI systems will gain competitive advantages in efficiency, regulatory compliance, and sustainability.
To implement these solutions, it is essential to have a technology partner that understands both knowledge engineering and artificial intelligence. Q2BSTUDIO combines both disciplines, offering from initial consulting to development and deployment in cloud environments. Additionally, its cybersecurity services ensure building data is protected, and its BI capabilities enable dashboards for management.
In summary, the complementarity between knowledge graphs and explainable AI is the key to effective and responsible urban mining. The technology is already available; it just needs to be applied wisely. Q2BSTUDIO is ready to lead that change, providing custom software applications that integrate AI, cloud, cybersecurity, and BI, all aimed at generating defensible and auditable decisions.
