Ontario Power Generation

To maximize plant reliability and availability, complex systems such as nuclear power plants (NPPs) continuously monitor and record the performance and health status of many components, equipment, and systems. Such data may take the form of online monitoring data, condition reports, or maintenance reports, and can provide systems engineers with insights into anomalous behaviors, degradation trends, and the possible causes behind each—in addition to enabling predictions as to their direct consequences. However, analyzing such large amounts of data poses challenges. While some of these challenges are technical in nature (i.e., data are often distributed over several physical servers or databases), others are conceptual (i.e., data elements come in different formats [numeric or textual]), and measured values have different scales (e.g., vibration spectra and oil temperature). This paper directly tackles these challenges, with a focus on integrating all these data elements to assist systems engineers in analyzing historic equipment and system reliability performance to support the optimization of maintenance activities. This integration is performed by extracting knowledge from textual data via technical language processing methods and by quantifying system, equipment, and component health from numeric condition-based data. We rely on model-based systems engineering (MBSE) models of systems and equipment to identify their architecture and functional (i.e., cause and effect) relations. Numeric and textual data elements are then associated with an MBSE graph element, based on their nature. This bonding of MBSE models and data elements constitutes a first of its kind knowledge graph of a NPP system, with data elements being organized in a structured manner that enables systems engineers to track historic reliability performance of systems/equipment (trends and patterns of events), identify cause-effect relations between data elements, and carry out appropriate actions in response.

Study Region: The Abitibi River watershed is a large hydropower-regulated basin located in northeastern Ontario and northwestern Quebec, Canada, where natural streamflow records are unavailable due to long-term flow regulation and reservoir operations. Study Focus: This study presents an uncertainty-aware framework for long-term streamflow reconstruction and trend analysis in a regulated, data-scarce watershed. The framework integrates hydrological modeling, machine-learning-based parameter estimation, and dendrohydrological validation to reconstruct natural streamflow in the Abitibi River watershed. Three machine learning algorithms are used to estimate parameters of the GR4J-CemaNeige model, generating pseudo-observations of natural streamflow. These pseudo-observations are then used to calibrate five conceptual hydrological models within the Raven platform, producing a 15-member ensemble of daily streamflow simulations for 1950–2023. New Hydrological Insights for the Region: The reconstructed ensemble shows strong agreement with flood-ring signature in tree rings of naturally flowing rivers, with significant correlations ranging from 0.35 to 0.72, and is further supported by comparison with scaled natural flows from a nearby watershed (annual CRPS = 49.39 m3/s). A hierarchical Bayesian linear model reveals statistically significant increases in annual and spring streamflow, with estimated trends of 1.91 and 2.67 m3/s/year, respectively. Overall, the framework provides a robust strategy for reconstructing long-term natural streamflow in regulated watersheds.

With the need to maintain long-term reliable energy using nuclear power plants, there is an underlying demand to ensure that the maintenance of plant components and systems is also done in an efficient and cost-effective manner. One way to achieve this is by moving from time-based maintenance to condition-based maintenance. The research presented in this paper focuses on applying statistical and machine learning-based methods to capture anomalies within data for fault detection to further develop into condition monitoring. This research focuses on system data for a circulating water system (CWS) of a pressurized heavy-water reactor for detecting anomalies. The different methodologies used for detecting and capturing anomalies in the CWS data are Matrix Profile, Density-based Spatial Clustering of Applications with Noise (DBSCAN), and Support Vector Machines (SVMs). Matrix Profile and DBSCAN are used to distinguish between normal data and anomalous data. This paper presents a hybrid method using DBSCAN and an SVM when a portion of the data is used for DBSCAN to generate clusters. This portion of data is then used to train the SVM along with the clusters generated by DBSCAN as output. The SVM is then tested on unseen data as a predictive tool, which can work in real time to categorize data points as either normal or anomalous. This paper presents results that show the high accuracies of DBSCAN and the SVM in capturing anomalies within the data for a CWS for fault detection. Thus, the maintenance plan would be focused on component condition rather than a time-based schedule by switching to an automated system to identify and predict faults within a CWS.

Energy utilities still run engineering work management, engineering procurement, and inventory processes on long-lived enterprise asset management platforms. Replacing these platforms is often cost prohibitive and operationally disruptive, so practical improvement layers are required. This paper presents a retrieval assistant that improves day-to-day knowledge access across three operational modes: vendor documentation question answering, operational data store (ODS) schema question answering, and user interface usage and how-to question answering. The runtime method combines intent understanding, query rewriting, hybrid semantic and vector retrieval, context engineering under token limits, grounded answer generation, and deterministic hyperlink conversion for panel identifiers and cited documentation. The data preparation pipeline emphasizes semantic enrichment as the primary quality lever by adding table and field descriptions, normalizing acronyms across sources, and indexing representative row-level context when useful. A measured pilot shows consistent gains in retrieval quality and user outcomes. Precision at five improved from 0.56 to 0.72, mean reciprocal rank from 0.43 to 0.58, and normalized discounted cumulative gain (nDCG) at five from 0.51 to 0.66. Median task completion time dropped from 14.2 to 8.3 minutes, while usefulness and confidence both increased to 4.0 on a five-point scale. Results are based on a small sample and are reported as pilot findings, but they indicate that intent understanding and semantic enrichment can deliver meaningful operational value in legacy environments while also establishing reusable foundations for future analytics and automation tools.

Retrieval-augmented generation (RAG) is the dominant paradigm for applying large language models (LLMs) to enterprise document corpora, yet naive implementations encounter hard limits as corpus scale and query complexity grow. This paper traces the evolution of a production retrieval pipeline at Ontario Power Generation (OPG) for regulatory compliance and rate case analysis under Ontario Energy Board (OEB) reporting requirements. We examine successive stages: naive RAG, hybrid retrieval with re-ranking, agentic function-calling retrieval, and a deep multi-agent architecture with code-based tool synthesis and explicit planning, and identify the failure modes and tradeoffs that motivated each transition. We formalize the mature architecture as Progressive Evidence Acquisition with Cost-Aware Escalation (PEA-CAE): begin with low-cost, high-precision retrieval and escalate to full-document reads only when the expected evidence gain justifies latency and cost. Our findings show that context engineering is a more tractable and economically viable path than domain-specific fine-tuning for large, evolving regulatory corpora. More broadly, the progression toward deep agentic retrieval mirrors classical information retrieval ideas, introducing adaptive query reformulation, progressive document discovery, and hierarchical subagent summarization as practical system primitives. Operational traces further support the search-based nature of modern retrieval systems, where iterative evidence acquisition and adaptive planning increasingly replace single-pass retrieval as the foundation for enterprise-scale question answering.