Introduction

The CANDU reactor—an acronym for CANada Deuterium Uranium—occupies a unique position within the global nuclear fleet as a pressurized heavy-water reactor design with distinct operational advantages. By employing heavy water (deuterium oxide) as both moderator and primary coolant, the CANDU design achieves a neutron economy efficient enough to burn natural uranium fuel without requiring enrichment. This property, combined with on-power refueling capability and a horizontal fuel channel configuration, grants CANDU stations an operational flexibility that light-water reactor designs cannot match. However, the distributed channel architecture that enables these benefits simultaneously creates a complex three-dimensional neutron flux distribution requiring exceptionally reliable core monitoring. Detecting subtle shifts in neutron behavior, coolant temperature profiles, structural deformation, or vibration signatures across hundreds of individual channels is essential—not only for optimizing fuel burnup but, more fundamentally, for preserving the integrity of the pressure-tube boundaries that serve as the primary safety barrier.

Over the past decade, a wave of technological innovation has begun reshaping how operators and engineers approach CANDU core surveillance. Traditional sensor suites are being augmented—and in select cases replaced—by fiber-optic networks engineered to withstand extreme radiation fields, solid-state neutron detectors offering microsecond response times, and machine-learning models capable of forecasting subtle anomalies long before they escalate into operational concerns. These emerging technologies are moving the industry away from periodic, offline analysis and toward continuous, predictive diagnostics that promise improved safety margins, reduced operational costs, and extended license life for aging CANDU units. This article examines the traditional monitoring toolkit, the limitations that spurred innovation, and the novel approaches now being deployed at CANDU sites across Canada and abroad.

Understanding CANDU Core Monitoring Challenges

At the heart of every CANDU reactor lies a large cylindrical calandria vessel housing hundreds of horizontal pressure tubes. Each tube contains a string of twelve or thirteen natural-uranium fuel bundles, separated from an outer calandria tube by a thin gas annulus. The moderator space between calandria tubes is filled with cool, low-pressure heavy water, while pressurized heavy-water coolant flows through the pressure tubes to remove heat generated by fission. This distinctive geometry introduces several monitoring challenges not encountered in single-vessel light-water reactor cores.

First, power distribution is inherently asymmetrical and evolves continuously due to on-power refueling—typically two to five bundles are exchanged daily in a large station. Each refueling operation causes localized flux tilts that must be tracked in near-real time to prevent fuel overheating or reactor power excursions. Second, because the heavy-water moderator is physically separated from the coolant, the dynamics of prompt neutron lifetime and delayed-neutron fraction behave differently under upset conditions compared to light-water designs. Third, mechanical degradation mechanisms—including creep, corrosion, deuterium ingress, and pressure-tube diametral expansion—develop slowly over decades, requiring sensitive monitoring of structural parameters that traditional electrical sensors struggle to capture with sufficient spatial resolution.

Ensuring safe and economical operation therefore demands a capable suite of in-core and ex-core instruments that measure neutron flux, temperature, coolant flow, pressure, and vibration—and increasingly, a data-fusion platform that transforms raw sensor signals into actionable intelligence for operators and engineers.

Traditional Monitoring Methods and Their Limitations

In-Core Neutron Flux Mapping

Since the 1970s, CANDU reactors have relied on self-powered neutron detectors (SPNDs) installed inside flux-detector guide tubes running axially alongside fuel channels. The most common type is the vanadium SPND—a small emitter wire of vanadium-51 that absorbs neutrons, then undergoes beta decay to produce a current proportional to local thermal-neutron flux. SPND arrays provide a spatial map of the core's flux distribution, feeding the station's Regional Overpower Protection Trip (ROPT) system and reactor-regulating computers.

Vanadium detectors are robust under intense radiation, but they exhibit an inherent time lag of several seconds due to the 3.7-minute half-life of vanadium-52. In fast-transient scenarios—such as a loss-of-coolant accident or sudden reactivity insertion—this lag can delay protective action. Cobalt-based SPNDs offer faster response but suffer from higher burnup rates and require more frequent calibration. Fission chambers, which provide instantaneous signals, see limited in-core use because of their larger size and shorter operational life in high-flux regions. Consequently, operators have historically complemented in-core SPNDs with ex-core ion chambers that deliver fast, spatially averaged flux readings for the entire reactor.

Temperature and Coolant Flow Monitoring

Thermocouples and resistance temperature detectors (RTDs) are positioned at fuel-channel outlets to measure coolant temperature rise, which correlates with channel power. When combined with flow measurements from venturi meters or ultrasonic probes on feeder pipes, thermal data provide an independent check on channel power. However, thermocouples degrade under neutron and gamma irradiation, and their calibration drifts require correction through periodic cross-comparison with gamma-based heat-output measurements. The point-wise nature of these sensors leaves significant gaps in the core's thermal profile, particularly during transient conditions where radial power distribution can shift rapidly.

Pressure-Tube and Feeder Inspection

Structural integrity assessment has traditionally been performed offline during scheduled outages. Eddy-current probes, ultrasonic thickness gauges, and remote visual tools inspect pressure tubes, calandria tubes, and feeder piping. While highly accurate, these periodic inspections cannot detect fast-onset degradation mechanisms, such as the sudden acceleration of delayed-hydride cracking. Consequently, station operators have long sought continuous, in-situ methods for monitoring the primary circuit's mechanical health.

The collective limitations of the traditional suite—slow response, coarse spatial sampling, calibration drift, and reliance on outage-based structural inspection—have created a clear opportunity for innovative technologies that deliver higher-fidelity data in real time.

Innovative Sensor Technologies for the CANDU Core

Fiber-Optic Sensors: Distributed Temperature and Strain Sensing

One of the most promising additions to the CANDU monitoring toolkit is the family of fiber-optic sensors. Silica-based optical fibers can be drawn through existing instrument tubes or bonded to the external surface of feeder pipes and calandria tube sheets, creating a nerve-like network that measures temperature and strain with millimeter-scale spatial resolution over distances of several hundred meters.

Distributed Temperature Sensing (DTS) uses the Raman scattering principle: when a laser pulse travels down the fiber, a portion of the backscattered light shifts in wavelength proportionally to local temperature. By measuring the time of flight of the returning pulse, the system maps temperature along the entire fiber length. In a CANDU context, a single DTS fiber can replace dozens of point thermocouples, providing a continuous thermal profile of the moderator, the annulus gas, or even the coolant outlet of every channel. Recent research at Canadian Nuclear Laboratories has demonstrated that pure-silica-core fibers can survive accumulated doses exceeding 100 MGy while maintaining a temperature resolution better than 0.5 °C, making them suitable for near-term deployment in the calandria environment.

Fiber Bragg Grating (FBG) sensors take advantage of periodic variations in the refractive index inscribed along the fiber core. Each grating reflects a narrow-band wavelength that shifts in response to strain or temperature. By multiplexing dozens of FBGs on a single fiber, operators can monitor minute deformations of pressure tubes, extension of annulus-gas flow paths, or distortion of calandria tube sheets. Unlike electrical strain gauges, FBGs are immune to electromagnetic interference—an important advantage inside the calandria, where high-voltage feeder cables generate substantial electrical noise. An illustrative field trial, reported in a Canadian Nuclear Laboratories technical summary, installed an FBG array on a decommissioned calandria mock-up and successfully detected sub-millimeter deflections under simulated thermal-cycled loads. Plans are now underway to deploy a similar system in a commercial unit during a planned outage, feeding strain data directly into the station's digital twin.

Advanced Neutron Detectors: Speed and Sensitivity

Solid-state neutron detectors are emerging as a compelling alternative to vanadium SPNDs for fast flux mapping. Two designs in particular have attracted interest: micro-pocket fission detectors and single-crystal diamond detectors.

Micro-pocket fission detectors consist of a tiny cavity—typically less than a cubic millimeter—filled with a fissionable gas such as 235U hexafluoride, surrounded by electrodes on a silicon substrate. When a neutron induces a fission event, the energetic fragments ionize the gas, producing a measurable current pulse within nanoseconds. Because the detector can be powered cyclically, its response is immediate compared with the several-second delay of a vanadium SPND. Fabricated using micro-electromechanical systems (MEMS) techniques, arrays of micro-pocket detectors can be placed along the entire length of a guide tube, delivering a high-resolution flux profile at every refueling step without the lag that limits fine-tuning of the ROPT system.

Diamond-based detectors exploit the wide bandgap and exceptional radiation hardness of synthetic single-crystal diamonds. When neutrons interact with a boron-10 or lithium-6 converter layer deposited on the diamond surface, the resulting alpha particles generate electron-hole pairs that drift to metallic contacts under bias voltage. The resulting current is linearly proportional to neutron flux with a response time well below one microsecond. In a 2022 study published by the IAEA's nuclear instrumentation program, diamond detectors maintained stable sensitivity after accumulating over 1017 neutrons/cm², far exceeding the lifetime of conventional fission chambers in the same environment. Both technologies offer a pathway to replace aging SPNDs and ionization chambers, reducing calibration drift while enabling digital signal processing that separates gamma-ray background from the neutron signal with higher fidelity.

Acoustic and Vibration Monitoring for Structural Diagnostics

Mechanical anomalies inside the primary heat-transport system often produce distinct acoustic signatures. The sudden movement of a loose part, the chatter of a vibrating feeder-pipe support, or the characteristic "ping" of a delayed-hydride crack initiation can be detected by piezoelectric accelerometers mounted on the outside of the calandria shell or on accessible feeder-pipe elbows. Long-range ultrasonic guided-wave techniques can also send a pulsed wave along a pressure tube and analyze reflections to detect wall thinning or crack-like defects without requiring an outage.

A multi-year project at Ontario Power Generation's Darlington station has installed a permanent network of accelerometers on the calandria face and on a subset of feeder pipes. Data are continuously streamed to a central processor where machine-learning algorithms classify events. The system has already demonstrated its ability to identify a drifting feeder-pipe support bracket, allowing a proactive repair during a planned maintenance window. Integrating these acoustic channels with neutron-flux and temperature data provides a holistic picture of the reactor's physical state—a concept often termed condition-based monitoring.

Intelligent Diagnostics: Analytics, Digital Twins, and Machine Learning

The influx of high-resolution sensor data from fiber-optic networks, solid-state neutron detectors, and acoustic arrays cannot be acted upon without an equally sophisticated data-analysis engine. CANDU operators are turning to machine learning and physics-based digital twins to fuse these disparate data streams into coherent, real-time diagnostics.

Machine Learning for Pattern Recognition

Supervised and unsupervised machine-learning models are being trained on historical station data to recognize the subtle precursors of known fault modes. For example, a neural network fed with five years of thermocouple, SPND, and flow data can learn the normal correlation patterns between channel-outlet temperature, flux tilts, and feeder-flow oscillations. When a new data point deviates from the learned pattern—say, a temperature rise in a channel without a corresponding flux increase—the system flags a potential blockage of coolant flow or a gauge calibration error. These models are especially effective at detecting slow-onset phenomena that might escape a human operator until they breach an alarm threshold.

Deep-learning autoencoders have been applied to frequency-domain analysis of neutron-flux signals. In a collaborative study between the University of Ontario Institute of Technology and Bruce Power, researchers trained an autoencoder on normal core-behavior frequency spectra derived from in-core SPND data. The reconstruction error increased noticeably in the weeks before a channel's flow-degradation event, giving operators an early-warning window of several days.

Digital Twin Technology

Whereas a traditional simulator models the reactor's neutronics and thermal-hydraulics using nominal design parameters, a digital twin is a living model continuously updated with live sensor data. By feeding real-time flux maps, temperature profiles, and mechanical strain readings into a high-fidelity core-physics code such as DYN3D-GANDU or R-FIT, the twin mirrors the actual core state second by second. If a sensor reports a local flux increase that cannot be explained by refueling, the twin can run a forward simulation to predict the resulting channel-power evolution and recommend either a power reduction or an adjustment to the liquid-zone-control absorber levels.

The long-term vision, articulated in the Canadian nuclear industry's Digital Twin Roadmap, is to couple the twin to an automated decision-support system that proposes corrective actions to the licensed operator, thereby reducing human cognitive load during transients. In the next generation of CANDU refurbishments—such as the ongoing Bruce Major Component Replacement project—digital twins are being integrated into the new I&C architecture from the start.

Edge Computing and Data Fusion

With fiber-optic interrogators generating gigabytes of raw data daily, moving all processing to a central control room is impractical. Emerging architectures deploy edge computing nodes directly on the sensor networks. These nodes perform preliminary noise filtering, event detection, and data compression before transmitting only actionable data to the digital twin. FPGAs and embedded processors hardened for radiation environments are now being evaluated for this role. This distributed intelligence reduces bandwidth requirements and ensures that critical alarms are not delayed by network latency. The integration of edge analytics with the station's existing distributed control system represents a key enabler for widespread adoption of high-density sensor arrays.

Real-World Pilots and Proven Gains

Several CANDU sites have moved beyond laboratory prototypes to full-scale field deployments. At Point Lepreau NGS, a DTS fiber was routed through a spare instrument thimble to measure moderator temperature during a startup; the data revealed a previously unidentified cool spot near a calandria-tube-sheet weldment, prompting a targeted ultrasonic inspection that confirmed the onset of creep. The early catch avoided a potential tube-sheet leak and saved an estimated two months of outage time.

At Ontario Power Generation's Pickering station, a trial of micro-pocket fission detectors in a single refueling channel demonstrated that fast-response flux signals could reduce the uncertainty in the ROPT detection time by nearly 40 %, a safety margin improvement that may permit a modest power uprate in the plant's final operating years. These pilots underscore that the innovation is not merely theoretical; it is delivering measurable operational and safety benefits today.

Integration with Plant I&C Systems

The success of pilot projects hinges on seamless integration with the station's existing instrumentation and control (I&C) architecture. At Bruce Power, engineers have developed a middleware layer that translates raw sensor data from fiber-optic and acoustic networks into standard MODBUS and OPC UA formats, allowing the digital twin to ingest data without modifying the legacy trip systems. This approach preserves regulatory qualification of safety-critical equipment while enabling advanced diagnostics on the non-safety side. Such integration strategies are essential for retrofitting older CANDU units without requiring a complete I&C overhaul.

Hurdles on the Path to Adoption

Despite their promise, the new monitoring technologies must navigate a conservative regulatory environment and practical engineering constraints. Radiation hardness remains the central challenge. Even radiation-tolerant optical fibers develop color-center absorption that raises attenuation over time, requiring careful material selection and periodic recalibration. Component qualification for the harsh in-core environment—where temperatures can exceed 300 °C—demands extensive testing, often taking years before a device receives regulatory clearance.

Data overload is another concern. A single fiber-optic DTS interrogator can generate several gigabytes of raw data per day. Without robust compression and on-board preprocessing, the data-acquisition system can become a bottleneck. Cybersecurity is also paramount; as core-monitoring systems become more interconnected, they must adhere to the strict guidelines of IEC 62645 for digital I&C security, ensuring that diagnostic signals cannot be spoofed or disrupted. Addressing these non-technical barriers is as critical as perfecting the sensors themselves.

Furthermore, the cost of retrofitting existing stations with advanced sensor networks must be justified against the value of improved safety margins and reduced outage durations. Utilities are conducting cost-benefit analyses that factor in avoided forced outages, extended fuel cycles, and regulatory credit for enhanced monitoring. Early adopters have found that the return on investment is compelling, especially for units approaching the end of their design life where every outage day saved translates directly into revenue.

Future Directions: Toward a Self-Diagnosing Core

Looking ahead, the convergence of advanced sensors, artificial intelligence, and secure I&C platforms points toward a future in which the CANDU core becomes largely self-diagnosing. Research teams at Canadian Nuclear Laboratories and academic partners are exploring the integration of augmented reality interfaces that overlay real-time sensor data onto a 3D core model, allowing operators to "walk through" the reactor and see temperature, flux, and vibration fields represented as color maps. In the same vein, physics-informed neural networks are being trained to solve the neutron-diffusion equation directly from sensor data, effectively merging model-based and data-driven approaches into a single diagnostic tool.

Longer-term efforts focus on autonomous diagnostic systems that can optimize reactor performance—adjusting refueling rates, liquid-zone compensator levels, and even safety-system setpoints dynamically—while keeping every operating parameter within the safe operating envelope. With several CANDU units now expected to operate beyond 60 years through life-extension programs, there is a powerful economic incentive to adopt monitoring systems that can detect aging phenomena early, guide targeted refurbishment, and demonstrate continued fitness-for-service to the regulator. The ultimate vision is a fully instrumented core that reports its own health continuously, empowering operators to make proactive decisions rather than reacting to events.

Conclusion

The transformation of CANDU core monitoring from a largely reactive, point-sensor-based paradigm to a predictive, network-aware model is well underway. Fiber-optic sensors, fast-response neutron detectors, and acoustic condition-monitoring networks are providing a data-rich foundation, while machine-learning analytics and digital twins extract actionable foresight from that data. These innovations are not merely incremental improvements; they represent a fundamental shift in how nuclear operators understand and safeguard the heart of the reactor. As pilot projects mature into fleet-wide deployments, the CANDU community stands poised to achieve levels of safety, reliability, and economic performance that were unimaginable even a decade ago.