Multi-modal fiber optic monitoring system for transformer failure detection and lifetime estimation

Abstract

A multi-modal transformer monitoring system integrating optical fiber sensors for acoustic, chemical, and temperature measurements. The system can monitor electrical parameters such as voltage and / or current. Additional chemical and physical parameters may be measured, such as humidity, moisture, pressure, and others. A signal interrogation module digitizes and conditions the sensor output, and a sensor fusion engine applies machine learning and adaptive filtering to extract features and detect transformer failures such as partial discharges, thermal anomalies, and oil degradation.

Description

TITLEMulti-Modal Fiber Optic Monitoring System for Transformer Failure Detection and Lifetime EstimationCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit under 35 USC § 119 of U.S. Provisional Application No. 63 / 789,173, filed on April 15, 2025, which is incorporated herein by reference.STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH

[0002] This invention was made with United States government support under DE-SC-0025043, DE-SC-0025125, and DE-SC-0025055 awarded by the Department of Energy.BACKGROUND OF THE INVENTION

[0003] The present disclosure generally relates to relates to transformer health monitoring systems. More specifically, the disclosure relates to integrating multi-modal sensing technologies and advanced data analytics, including artificial intelligence enabled, to detect and diagnose incipient failure events in power transformers.

[0004] Power transformers are critical components in electrical grids, and their unexpected failures can lead to significant operational disruptions and economic losses. Electric power utilities are facing increasing challenges to maintain reliability in territories due to increasing demand combined with increased penetration of renewables, electric vehicles, and other distributed generation resources on the distribution system. Distribution transformers represent a critical link between the medium voltage electrical grid infrastructure and consumer, playing a crucial role in energy efficiency and grid reliability. Traditional monitoring techniques often rely on single-parameter assessments, which may not provide comprehensive insights into transformer health. Other sensing solutions would not be possible with the medium and high voltage transformers. Therefore, it would be advantageous to develop an integrated system capable of multi-modal sensing and advanced data analysis to enhance transformer diagnostics and maintenance strategies.BRIEF SUMMARY

[0005] According to embodiments of the present disclosure is a multi-modal transformer monitoring system integrating optical fiber sensors for acoustic, chemical, and / or temperature measurements. In some embodiments, electrical parameter monitoring such as voltage and / orcurrent is also possible. A signal interrogation module digitizes and conditions the sensor output, and a sensor fusion engine applies machine learning and adaptive filtering to extract features and detect transformer failures such as partial discharges, thermal anomalies, and insulation / oil degradation. The system provides real-time health diagnostics including instantaneous or cumulative loading, incipient failure detection, life expectancy predictions, and fault classification.

[0006] In one embodiment, the system uses fiber optic sensing integrated with additional low-cost sensors to detect transformer incipient failure events. It integrates one or more of acoustic, chemical, temperature, current and voltage sensing to diagnose key failure modes (e.g., partial discharges, mechanical vibrations, oil dissolved gases, thermal anomalies, electrical faults) and thus improve transformer health monitoring by lifetime estimation through the post-processed sensing data via an advanced non-linear system control algorithm. It includes communication options such as wireless and also options for energy harvesting technology to avoid the need for active power at the sensor.BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0007] Fig. 1 A an overview of the system according to a one embodiment.

[0008] Fig. IB is a schematic diagram of components of the system.

[0009] Fig. 2A is a flowchart of the data processing and fusion according to one embodiment.

[0010] Fig. 2B is a flowchart of the data processing and event detection process.

[0011] Fig. 3 shows the optical fiber sensor deployment within a transformer.

[0012] Fig. 4 depicts an interrogation module.

[0013] Fig. 5 shows the multi-channel signal conditioning in an interrogation module.DETAILED DESCRIPTION

[0014] According to embodiments of the disclosure is a monitoring system 100 comprising optical fiber-based sensors 101 installed in or around a transformer to monitor various parameters including acoustic emissions, gas content, and / or temperature. The system 100, according to one embodiment, is shown in Fig. 1A, and includes a optical fiber sensor 101, an energy harvester circuit 120, and a signal processing circuit 113 as part of the core components. The system 100 uses a photonic interrogation module 110, which is shown in Fig. IB, to illuminate the sensors 101 and a detector array 112 to read the sensor output, where the processor 113 analyzes the signals produced by the sensor 101. In some embodiments, thefiber optic sensors 101 are combined with additional sensors to provide complementary information. The interrogator 110 may be comprised of one or multiple sensors 101 and detection circuits 112, with multiple sensors 101 and detection circuits 112 enabling multiparameter functionality and sensor self-referencing to avoid impacts of opto-electronics fluctuations, environmental effects, and aging on sensor outputs and functionality. In some embodiments, electrical parameter monitoring such as voltage and / or current is also possible. Additional embodiments may also monitor additional physical or chemical parameters including pressure, humidity, moisture, and others.

[0015] By way of further detail, the system 100 is a multi-modal monitoring system that uses fiber optic sensors 101 integrated with additional low-cost sensors 105 to detect transformer incipient failure events. The system 100 integrates acoustic, chemical, temperature, current and voltage sensing to diagnose key failure modes (e.g., partial discharges, mechanical vibrations, oil dissolved gases, thermal anomalies, etc.) and thus improve transformer health monitoring by lifetime estimation as well as risk of failure and asset loading amongst others, through the post-processed sensing data via an advanced non-linear system controllers 115.

[0016] The sensor 101 is a fiber optic-based device coupled with a sensor interrogator 110 that is self-powered with the secondary voltage of the distribution line transformers being monitored. In one embodiment, the system 100 utilizes a multimode fiber sensor device. This power is harvested using a current-transformer (CT)-based energy harvester circuitry 120. The design is intended to enable multi-channel sensor fusion, monitoring real-time heat rise data and can also be used for monitoring of line power quality through AC line harmonics, potentially measured in acoustics frequency spectrum. An example heat sensor 101 could be a thermo-optic thin-film-based fiber sensor operating at a wavelength of 400-600 nm and an optical power of 10-540 pW. An example acoustic sensor 101 could be a low-cost multimode interferometer operating at a wavelength of 800-850 nm or 1310-1600 nm and an optical power of 1-20 mW. Distribution transformers are known to operate at 50 to 65 °C under normal operation conditions and can rise to 80 °C or higher due to load anomalies or insulation degradation / aging. By utilizing a real-time sensor 101, an abrupt heat rise can be immediately followed by a trip-off alarm temperature established by a given original equipment manufacturer or utility, such as a limit set at 85 - 90 °C.

[0017] Thermal and chemical fiber optic sensors 101 for example can utilize evanescent wave / leaky-mode / reflection-mode intensity -based multimode fiber sensing mechanisms which enables low-cost excitation and detection optics in the optical interrogation modules 110 that are more cost-effective than commercial fiber optic sensing technologies. Additional to thevalue of the information provided by each sensor 101 independently, combined temperature and chemical sensing enables advanced schemes such as real-time signal compensation to correct for the temperature fluctuations in the gas sensing signals that are otherwise a pain point for existing fiber optic-based gas sensors. Example fiber optic acoustic sensors 101 utilize multimode interferometer sensing mechanisms to enable low-cost optics, which is otherwise un-achievable by existing Fiber Bragg Grating based acoustic dynamic strain sensors.

[0018] For thermal sensors 101, the sensing elements can be fabricated with various thermo-optic coatings based on Au-NP nanocomposite materials with matrix films deposited on standard multimode fiber (MMF) using an evanescent wave (EW) configuration, as is known in the art. Low-cost acoustic sensors can be based on multimode interferometric fiber optic sensors with single-mode-multimode-multimode-single-mode (SMS) structure or a thin-core fiber (TCF) structure in the middle. The design and fabrication of the interrogation module 110 and power system 114 include illumination and distribution optics, signal control circuitry 113, energy harvesting circuitry 120, sensor and electronic enclosure and packaging.

[0019] In one example embodiment, FC / PC type connectors are used for the photodiodes and the lead fibers leading out to the sensor 101 and circulating back into the interrogation module 110. In one embodiment, a 1 x 4 multimode optical splitter features a 25% split of input light at each port. The monochromatic LED light source Ill is rated at 20 mW or above at maximum input current and with wavelength options from 500nm to 595 nm. The interrogator power management module 114 starts with a power switch that offers options for 120V step down to 5V from a wall outlet and energy harvester 120. Four transimpedance ampmlifiers (TIAs) are incorporated with three used for sensors and one for the LED light source 111 referencing. LED drivers with voltage capabilities ranging from 0 to 3.6V, and current capabilities 0.01 to 0.15 A. Each TIA has an offset voltage of 0.25 pV with a drift of 0.005 pV / °C. The input bias current is 0.4 nA at maximum, with a 3dB bandwidth of 10 MHz and 140 dB of common-mode rejection ratio (CMRR). The current capability per channel is 125 mA. For signal sampling after the TIA voltage output, an example can be 10-bit analog-to-digital-converter (ADC) with the range of amplification to 0.5 to 1.5 V. The total current requirement for the entire board is 1.2 A, with a total voltage limited to 2.5 minimum and 18 V maximum. The final output powers the interrogation module 110 while ensuring system safety and stability through integrated feedback mechanisms. The energy harvester system 120 comprises several components, including an energy source, and AC / DC converter, overvoltage protection, a supercapacitor, and boost conversion.

[0020] To ensure compatibility with the fiber-optic sensor interrogator 110, the harvested power is conditioned using a boost converter, which steps up the voltage to the required levels. Since stable and noise-free power is crucial for accurate sensing, a low-noise RF linear regulator can be incorporated to further refine the output, reducing fluctuations that could interfere with sensor measurements. The boost conversion stage utilizes an inductor-based step-up topology to increase voltage. The feedback control loop ensures consistent output, preventing variations from interfering with the sensor interrogator’s operation. Further, an analog to digital converter can be used to operate the system 100 at a higher frequency.

[0021] Within the optical signal interrogation module 110, self-referencing of light source optical power is adapted with an additional optical path, enabling long-term power drift correction of the baseline optical power level of the sensor signals in the post-processing steps. The incorporation of existing electrical -based voltage and current sensors 105 and the integration of electrical-optical sensing technologies in a modular fashion enable hardware fusion, accomplished by a sensor fusion engine running on the controller 115.

[0022] The sensor fusion engine may comprise software, firmware, application specific hardware, or a combination of the foregoing. In one embodiment, the sensor fusion process is implemented on a controller 115. In this example embodiment, the sensor fusion process starts with calibrated sensor data using Neural Network algorithms, for example, long-short term memory (LSTM) or time delay neural network (TDNN), into parameterized readouts, then post-processed using a non-linear state estimation method including recursive and / or adaptive filtering algorithms. New variables as “states” are generated for user information about the transformer’ s health condition. These signal processors then extract relevant features, establish decision boundaries, and estimate the probability of false positives to reliably distinguish abnormal events from normal operation.

[0023] The sensor fusion engine can be used to extract actionable intelligence at the edge from thermal and acoustic data. The derived metrics enable prediction of transformer incipient failure and remaining life. Specifically, thermal metrics Top Oil Temperature (TOT) and Hot Spot Temperature (HST) led to Remaining Useful Life (RUL) prediction. Similarly, acoustic analysis of low frequency vibrations from the transformer core and winding lead to prediction of transformer wear and tear. Furthermore, analysis of high frequency acoustic signatures enable partial discharge monitoring.

[0024] Fig. 2A shows an example embodiment of the sensor fusion engine running on the controller 115. The flowchart depicted in Fig. 2 A shows the cascaded architecture of the exemplary embodiment of a Kalman Filter Based State Estimation algorithm used to predictthe RUL of the transformer. The transformer lifetime is a function of the dielectric breakdown of the insulation within the transformer, which can be based on IEEE standards. The process expands upon Arrhenius equation based quantitative method in the IEEE standard to provide a stochastically optimal estimate accounting for inherent noise in the measurement of the temperature sensing using Kalman filters. Kalman Filter is a robust state estimator that assumes the transformer's physical process as a sequential gaussian process. The lower filter in the cascade assumes thermal equivalent circuit as the dynamics of the modelling the relationship between TOT, current, and HST. Whereas the upper filter takes the predicted HST from the lower level as input to predict the RUL of the transformer based on Arrhenius rate theory. Variations of Kalman Filter like Extended Kalman Filter (EKF) were implemented to model the non-linear nature of the transformer’s Thermal equivalent circuit. The sensor fusion engine in this example can be written in Python, or alternatively, a resource efficient form in C++ and deployed in an embedded setup, such as ESP32. The sensor fusion engine is capable of operating without significant lag in ESP32, which further proves the effectiveness of the architecture as an edge analytics system.

[0025] Fig. 2B shows additional details about the sensor fusion and detection process, which enables multi-variable analysis, improving diagnostic accuracy. For example:

[0026] Acoustic + Chemical = Detect and verify partial discharges.

[0027] Temperature + Chemical = Link overheating with insulation degradation and / or oil degradation.

[0028] Acoustic + Temperature = Correlate vibration or arcing with local heating, mechanical deformations, and structural failures.

[0029] Acoustic + Temperature + Chemical = Loose connections, contact degradation, onload tap changer malfunction.

[0030] Temperature + Current = Enhanced lifetime prediction and real-time stress.

[0031] Current + Voltage = Local power quality and transformer efficiency.

[0032] Voltage + Current + Chemical + Temperature = Overloading or load-driven thermal stress.

[0033] The fusion engine outputs remaining life estimates and triggers maintenance alerts.

[0034] While the RUL accounts for degradation of transformer insulation, it fails to account for mechanical wear and tear of the transformer. The system 100 can include a method to track such degradation using acoustic signals utilizing an FFT algorithm to calculate the Total Harmonic Distortion in the signal. Basically, the vibroacoustic signals of the transformer are caused by core and winding vibrations (usually 120Hz for 60Hz input), which is a functionof input frequency signal. The presence of higher order harmonics of this signal indicates winding and core faults.

[0035] Similarly, partial discharge is another metric that has strong correlation to transformer health. Increase in occurrence of partial discharge indicates degradation of transformer health. As one example, partial discharge can be detected in the signal by utilizing a machine learning based method where a signal is converted to a spectrogram. The spectrogram acts like a digital footprint of the transformer which can be used to train a vision based (CNN) model to recognize the pattern associated with partial discharge. This process of detection is robust and does not depend on handpicked features like thresholding in previous methods. Additional methods include Fast Fourier Transforms and feature extraction.

[0036] RUL and THD provide health monitoring and incipient failure detection capabilities at the edge. LoRa is an example of a wireless network protocol and technology that can also be used for communicating these metrics and sensor measurements, with potential for additional and alternative to also be utilized. Based on actionable intelligence at the asset level, the grid controller module (ex. SCADA) can perform downstream analytics with predictive maintenance and fault isolation. Conventional methods focus on periodic analysis of inspection-based measurements for predicting maintenance needs. However, with RUL the system 100 can estimate transformer deterioration in real-time and maintenance becomes an event driven task wherein the monitored asset provides alerts to trigger the request for maintenance in addition to planned maintenance. This leads to cost savings for the grid operator and customer avoiding delays and costly failures. Additionally, the information can be integrated with advanced grid monitoring and control methods such as demand response, where the electrical loading of the asset is actively managed by the utility or a consumer to maximize lifetime and minimize risks of failure

[0037] Similarly, THD and RUL both can be utilized as metrics for incipient failure detection which makes the isolation of faulty asset predictive rather than reactive. This enables avoidance of further damage through fault propagation across the grid network through integration with broader grid analytics.

[0038] In addition to the metrics discussed above, analyzing the gas concentrations can be used to detect and classify faults. Dissolved gas analysis has shown to be a good indicator of a transformer’s insulation health, and particularly the presence of carbon oxides and hydrogen gas. Additional physical and chemical parameters related to gas and liquid phases can also be monitored including humidity, moisture, and pressure. As such, a rich set of temporal data can also be used to estimate the RUL of the transformers identifying trends through advancedpattern recognition algorithms. Temperature data-based RUL can be fused with gas sensorbased RUL to gain higher confidence estimates of RUL. At the same time, all of this feature rich continuous data can be served to Grid Level SCADA systems through APIs via communication channels such as LORA. The SCADA systems can integrate the edge level sensing data and incipient failure signals for event-driven predictive maintenance isolating faulty assets significantly reducing outage risks. Additional edge level metrics like RUL and diagnosed faults will also enable SCAD As to detect any abnormalities at the grid level for early intrusion detection and prevention. Enhancements to grid analytics and applications can also be realized including demand response programs informed by asset loading, lifetime, and failure predictions to provide additional value to utilities and consumers.

[0039] In summary, the combination of two or more parameters from the multi-modal sensors system 100 informs and identifies different transformer failure modes, including high frequency partial discharge, low-frequency harmonic distortion, insulation aging / failure, oxygen leakage / oil contamination, winding faults, and other failures. The sensor-fusion-enabled identification of these failure scenarios in a real-time continuous monitoring scheme is not currently performed in existing transformer monitoring systems.

[0040] Referring again to Fig. 1 A, the system 100 according to this example embodiment comprises a core module 150 having a temperature-incorporated dissolved gas sensor suite, a temperature-incorporated acoustic sensor module as a first add-on accessory 151, and a current and voltage sensor module as a second add-on accessory 152. The two accessory sensor modules 151 / 152 are adapted to connect to the core module 150, such as through standard daisy-chaining configuration with a shared bus. Additional modular designs are also possible, and variations of the existing modular designs are possible as well. For example, the core module 150 could include humidity, moisture and / or other chemical parameters of interest that are based on coated sensing layers. As another example, pressure sensors can be an additional physical parameter embodiment of the temperature-incorporated acoustic sensor module 151 using similar multimode interferometry techniques.

[0041] The modular design of the sensor modules 150 / 151 / 153 allows for integration into the efficient power management and edge processing of sensor data. The core module 150, comprising a standard current transformer, energy harvesting system 120, and thermal / dissolved gas sensor 101 and interrogator 110, behaves as the central unit in which accessory sensor modules can be interfaced via external connections. An edge processor 160, located within the core module 150, receives real-time signals from each of the accessory modules 151 / 152 and assembles data packets for external data transmission. The signals fromthe first add-on accessory module 151 (temperature and acoustic) and the second add-on accessory module 152 (voltage and current) magnitude spectra are sampled and computed within their local respective signal processors 113 and / or controllers 115, reporting computed values to the edge processor 160. In some embodiments, not all features may be required. For example, energy harvesting or wireless communications may not be needed or desired in some transformer deployments.

[0042] Fig. 3 depicts a group of sensors 101 integrated into a transformer. The acoustic sensors 101 utilize devices such as multimode interferometers, Fabry Perot interferometers, and Fiber Bragg Gratings, and are configured to detect events such as core mechanical vibrations, inter-turn short circuits, and partial discharges. Chemical sensors 101 detect dissolved gases in transformer oil and / or in the headspace above the oil, indicating oil and insulation system degradation, aging, or contamination. Temperature sensors 101 identify hot spots and thermal anomalies such as cooling failures and winding overheating. In cases of electrical parameter monitoring, transformer loading and transients can be detected and monitored. Conventional sensors 105 can also be used in addition to the fiber optic sensors 101, such as current sensors and voltage sensors as well as environmental and ambient monitoring and external surface measurements of the transformer. In the example embodiment shown in Fig. 3, the system 100 employs three optical sensing modalities (acoustic, chemical, and temperature), with complementary current and voltage sensing, which are combined by the sensor fusion engine in the controller 115. Additional and / or alternative sensor modalities can also be integrated such as humidity, moisture, and pressure within the core modules as well as utilizing complementary sensors of other types. The fiber optic sensors 101 can be installed in the transformer despite medium and high voltages due to inherent electrical isolation as compared to traditional sensors.

[0043] By way of further detail, in one example embodiment, a fiber optic sensor 101 is inserted into a distribution transformer (25 kVA, 7200GY / 4160, 120 / 240V) through custom fiber fitting design and fabrication. The fitting - connector materials are made of Nylon and silicone both for compatibility with transformer internal environment and for minimization of stress exerted on fiber cables. The ports are evenly distributed radially around the fitting, which is necessary for pressure-tight and stress distribution. The internal fitting is made from a tin-cure silicone (Mold Max™ 60 from Smooth-on) using a molding 3D printer, while the external housing is machined and modified from an 14 NPT pipe fitting (5318N28, McMaster-CARR). During energization, a pressure build-up of 10 - 15 psi is expected from internal winding heat rise. Therefore, the fittings are adapted to maintain a pressure of up to 20 psi.

[0044] In this example, fiber sensor configuration may include four additional sensors 104 (two evanescent-wave low-cost sensors and two reflection FBG sensors as temperature references). The sensing location and fiber placement primarily considers hotspot temperature, which is the gap between the wound core and the winding in the middle of the wound core width, which is below insulation and between layers at about 6-10 mm on the primary side. In addition, top-oil temperature located at 2 inches below surface oil level is positioned. This localization sensing capability is unique with fiber optic sensors within transformer and enables high value information such as fault location identification and reducing detection time. The fiber fitting was able to hold the pressure build-up during the energization. Two acoustic sensors 101 based on different multimode interferometry mechanisms (SMS and TCF) are also deployed on the backside of the transformer external body to monitor the AC line frequency harmonics. In another example of deployment, the fiber acoustic sensors 101 can be inserted within the transformer for early detection and localization of partial discharge or winding faults using multi-point quasi -distributed interferometer sensor elements.

[0045] By locating sensors at specific locations within the transformer, information content is enhanced and can further perform at least one of the following: measure acoustic emission at multiple locations for the purpose of localizing events in addition to detection, including partial discharge; measure temperature at high value locations, such as the hot spot or within the windings, for the purpose of providing high value temperature information; measure temperature at multiple locations to provide information about thermal distribution and heat flow throughout the transformer; and measure chemical parameters at specific locations in which evolved gases will initiate early in the degradation process to provide early indicators of health.

[0046] The system’s sensor and interrogator module operation can be validated within the energized transformer and demonstrates acquisition of sensor measurements that are able to be taken during transformer operation. The system 100 measured the optical feedback response of the sensing fibers driven by the interrogation module 110, which were installed manually into the transformer’s mineral oil. Oil samples can be taken from both transformers for both DGA and Standard Quality (SQ) Bundle to validate the intact quality of transformer oil after fiber optic sensor insertion. Then, the instrumented transformer is energized from the secondary voltage (backfeeding) with 208V (< 20 A) and expecting 7200 V floating on the primary (energized for total ~ 45 mins). The interrogation module 110 achieved reading baseline measurements wirelessly detecting minimal temperature change occurring in the duration of the test, maintaining relative stability throughout duration of the test. The low-costinterferometer fiber acoustic sensor 101 and custom interrogation module 110 was able to successfully capture the first and second harmonics during the period of energization.

[0047] In addition, successful validation of a high frequency bandwidth acoustic sensor 101 based on multimode interferometry can be demonstrated on a medium voltage power transformer. In this demonstration, a 5MVA, 23kV / 13.8kV, 60 Hz, three-phase, pad-mounted transformer within a modular integrated transportable substation (Eaton MITS), was integrated with two highly sensitive fiber acoustic sensors 101 retrofitted on the surface. The fiber sensors 101 are strategically placed on the transformer to capture acoustic emissions from various locations. Vibration signals collected from two fiber sensors 101 were analyzed by remote connected data acquisition. Measured time-domain signals were transformed using fast Fourier transform (FFT) to obtain a frequency-domain spectrum. During normal operation, transformer internal structures produce periodic vibration with harmonic signatures and stable frequency spectral response. In case of faults and failures (slackness of windings and cores, breaker failure, etc.), vibration features and frequency spectral responses will alter providing opportunities for real-time diagnostics. Finally, with the standard DGA tests and SQ bundle tests on the oil samples of both the fiber-instrumented and uninstrumented distribution transformers, the system 100 has been verified that the embedded fiber optic sensors 101 pass the certifications following IEEE C57.106-2015 for moisture and acidity levels and C57.104-2019 for combustible gas levels. All tests have passed the minimum requirements certified as Class-I oil conditions suitable for continued service.

[0048] The sensors 101 can be used for acoustic, pressure, moisture, humidity, gases, and temperature sensing. For acoustic sensing, the sensors 101 may include multimode interferometers (point and / or arrays) to passively capture acoustic signatures, such as vibrations from core mechanical damage, impacts from inter-turn short circuits, signals from corona discharge, and power line frequency anomalies, at high sampling frequencies (1kHz to 500kHz or greater). In case of multimode interferometers, the light source 111 may include low-cost Vertical Cavity Surface Emitting Lasers (VCSELs) or Light Emitting Diodes (LEDs). Detector arrays 112 can include silicon diode (or their arrays) with custom transimpedance amplification. The subsequent analog-to-digital converter (ADC) may be equipped with sampling of 2 MHz and / or above, and 16-bit resolution. Fig. 4 shows a sensor 101 with a light source 111 and detector 112 containing within an interrogation module 110.

[0049] Fig. 4 shows the interrogator, according to one embodiment. The interrogator integrates at least one light source, detector, signal converter / processing units, and power management. An optical fiber guides light from the light source to the sensing region. Theinterrogator can integrate multiple channels, with the ability to add additional interrogators. In many cases, the sensing region is a specific section of the fiber that is “sensitized” and responsive to environmental changes through internal or external functionalization. More specifically, the interrogator module combines all excitation sources, detectors, demodulators and the analog sensing / conditioning front-ends and the digitization of the associated optical and electrical sensors onto a signal integrated board. Power management can further include a battery and battery backup integration into the power management system. Fig. 5 shows the signal processing of the interrogation module.

[0050] For chemical sensing, the sensor 101 may comprise an array of evanescent wave / end-reflective fiber optic gas sensors that analyzes the transformer oil for dissolved gas markers (hydrogen, carbon dioxide, methane, carbon monoxide, acetylene, and similar compounds) indicative of oil or insulation material degradation, sludge formation, and oil leaks. These sensors 101 can be humidity protected with packaging such as porosity engineered Teflon or related fluorinated polymer optical capping layers. Additionally, the sensors 101 may be designed and selected to detect and monitor humidity, moisture, or other chemical parameters. Each of the gas sensors 101 itself can be made selective to individual gas species by, for example, a nanostructured thin film that exhibits optimal response at the relevant operating temperatures without temperature controlling features. In some embodiments, porous sorption-based materials and / or polymers can be utilized for gas, humidity, or moisture sensing. In some embodiments, combinations of sensors 101 are used to improve selectivity and / or sensitivity to one or more analytes. The multiple fiber array module forms an optical fiber photonic nose system that response to different gases at different wavelengths. These responses are excited by a low-cost broadband LED source (for example, 470 to 850 nm) and then de-multiplexed using an approach such as an array waveguide grating, and finally capture each signal with a detector, such as an array of silicon pixel detector chips.

[0051] For temperature sensing, the sensor 101 may comprise an embedded evanescent wave / end-reflection-based fiber optic sensor, with thermo-optic coating to monitor localized temperature changes to detect top-oil temperature variations, cooling system failures, and overheating in windings or core areas potentially even hot spot temperatures directly. In one embodiment, the temperature sensor 101 can also be the same low-cost interferometric sensor as the acoustic sensor, with signal discrimination approach such discrimination matrices. This is also an integrated piece to the “photonic nose” chemical sensing system mentioned above that enables real-time temperature compensation of the fluctuating gas sensing response due to continuously varying temperature and it also provides critical information to the transformerremaining useful lifetime, risk assessment, and incipient failure indicators. The ability to position the sensor 101 at numerous locations within the transformer enabled by the flexibility and isolation capability of fiber optic cables is a unique advantage and can even enable a concept of a “photonic nervous system” where measurements are made at high value locations and can be single parameter or multiparameter in nature. The location of deployment of the sensor 101, such as at the hot spot and within the winding or near the insulation materials, can enhance and diversify the sensing information that is obtained which can also provide additional information and value by using advanced signal and data processing schemes.

[0052] Referring again to the sensor fusion engine, data acquired are first filtered and drift corrected (opto- electronic drift) using adaptive filtering. The filtered signals are then calibrated using artificial intelligence processes, such as time delay neural network, into parameterized readouts. This calibration process can involve drift compensation (sensor materials drift), field transient response correction, and parameterization using prior field tests dataset. Then, they are post-processed using non-linear state estimation method including adaptive filtering algorithm. New variables as “states” are generated for user information. The processes then extract relevant features, establish decision boundaries, and estimate the probability of false positives to reliably distinguish abnormal events from normal operation.

[0053] Current and voltage sensing can be achieved using engineered fiber optic sensing approaches to mitigate technical challenges with medium and high voltage integration, or alternatively with more traditional current and voltage sensors available. Combination of current and / or voltage sensors with the sensors described above provides unique value in terms of transformer loading in addition to health monitoring. Example current sensors using fiber optics may include integration of fiber optics with magneto-optical materials in form of thin films, single or polycrystalline bulk samples, or composite materials such as ferrofluids or polymer - magnetic nanoparticle composites. These materials can be integrated with similar interrogation systems as described above, or alternatively using novel polarization dependent interrogation schemes

[0054] Light from sources such as broadband light emitting diodes (LEDs) with tailored phosphor materials or Vertical Cavity Surface Emitting Lasers (VCSELs) excites the sensors, and photodetectors capture the response. Analog signals are conditioned, digitized at high resolution (e.g., 20-bit, 2-10 MHz ADC), and processed through an integrated circuit. In some embodiments, photonic integrated circuits are utilized. Data processing enables processing of raw signals such as quality assurance, denoising, referencing to additional signals, filtering, smoothing, and interpolation as required. In some embodiments, the data processor alsoexecutes models such as machine learning (ML), artificial intelligence (Al) and related as well as adaptive filters to extract signal features and assess health metrics.

[0055] As described herein, the system provides: (1) Opto-electronic integration, including embodiments which are on-chip such as integrated photonics and photonic integrated circuits, to expand sensing capabilities and sensor fusion. (2) Highly localized chemical sensing with fine area resolution for fast response to dissolved gas species at affordable price point at the distribution level. (3) Low-cost acoustic sensing using on-device interrogation. (4) Low powerconsuming, edge-computed state-of-health estimation of power transformers in real-time. (5) Enhanced data fusion minimizes false positives and provides a more detailed state-of-health assessment.

[0056] The system 100 can be implemented as follows: 1. Sensor Installation: A specialized optical fiber is installed along / within the transformer environment, with defined “sensitized” regions for environmental responsiveness. 2. Excitation: Optical intensity irradiation from an emission source (e.g. LED or Laser) and guided by optical fibers. 3. Signal Generation: (i) Acoustic sensors generate intensity modulation from the captured vibrations and acoustic emission signatures, (ii) Chemical sensors generate optical intensity changes from optical constant changes of the sensing thin film in response to adsorbed chemicals in the oils, (iii) Temperature sensors generate optical intensity changes from thermo-optic effect of the refractive index change in the sensing layers or the fiber cavity, (iv) In some cases, current and or voltage sensors can also generate intensity modulation from magneto-optical or electro-optical responses of materials integrated with the fiber, (v) In additional cases, additional chemical and physical parameters may be measured such as humidity or moisture through modified optical constant changes of sorbent materials, thin films or polymers, and pressure through impacts of pressure on sensing materials or diaphragm structures such as a Fabry Perot interferometer. 4. Signal Acquisition & Conversion: Generated signals are detected by photodiodes as photocurrents and converted to analog voltage signals through operational amplifiers. 5. Signal Conditioning & Transmission: The analog voltages are then conditioned for integrity through intermediate integrated circuits before being digitized by analog-to-digital converters and finally transmitted to a central processor / controller. 6. Data Processing & State Variable Estimation: The processing module processes the signals by first calibrating them via analytics algorithms, including the utilization of artificial intelligence in some embodiments, into labeled parametric data, then uses adaptive filtering to estimate state variables for lifetime predictions. Data processing can further include self-referencing, filtering, smoothing, denoising, and similar signal processing steps. Filtering may also include taking the raw signal,applying self-referencing and sensor channel normalization (i.e. Savitzky-Golay smoothing, reference channel normalization, self-reference correction, sensor channel normalization), outlier suppression and dual-stage filtering (Hampel filtering and Kalman filtering) to arrive at a filtered signal. Finally it extracts features and sets decision boundaries to determine abnormal events. 7. Event Reporting: Detected events are logged and reported, providing actionable insights regarding transformer state-of-health. In addition to specific events, life estimation, real-time transformer stresses, and incipient failure probabilities and alerts are also enabled. While the method refers to transformers, the system can be utilized with other equipment, industrial products, hardware, and similar devices.

[0057] When used in this specification and claims, the terms "comprises" and "comprising" and variations thereof mean that the specified features, steps, or integers are included. The terms are not to be interpreted to exclude the presence of other features, steps or components.

[0058] The invention may also broadly consist in the parts, elements, steps, examples and / or features referred to or indicated in the specification individually or collectively in any and all combinations of two or more said parts, elements, steps, examples and / or features. In particular, one or more features in any of the embodiments described herein may be combined with one or more features from any other embodiment s) described herein.

[0059] Protection may be sought for any features disclosed in any one or more published documents referenced herein in combination with the present disclosure. Although certain example embodiments of the invention have been described, the scope of the appended claims is not intended to be limited solely to these embodiments. The claims are to be construed literally, purposively, and / or to encompass equivalents.

Claims

CLAIMSWhat is claimed is:

1. A multi-modal transformer health monitoring system, comprising:a plurality of multiplexed optical fiber sensors configured to detect acoustic emissions, chemical species, temperature fluctuations, humidity, pressure, moisture, and / or electrical parameters within a transformer environment;an interrogation module comprising at least one light source, one detector array, and circuitry for analog and digital signal processing; anda sensor fusion engine configured to accomplish at least one of the following:classify events using acoustic, chemical, thermal, physical, and / or electrical data, diagnose transformer faults including partial discharges, oil degradation, thermal anomalies, and incipient failure,estimate transformer state-of-health and remaining useful lifetime using machine learning and adaptive filtering algorithms, and.identify other performance characteristics.

2. The system of claim 1, wherein the plurality of multiplexed optical fiber sensors comprise acoustic sensors configured to detect specific failure modes selected from a group consisting of: core mechanical damage, inter-tum short circuits, corona discharge, and power-frequency anomalies.

3. The system of claim 1, wherein the plurality of multiplexed optical fiber sensors comprise chemical sensors configured to identify failure modes selected from the group consisting of: oil moisture contamination, sludge formation, aging and degradation of insulation, oil leaks, and corona-related gas emissions.

4. The system of claim 1, wherein the plurality of multiplexed optical fiber sensors comprise temperature sensors configured to detect thermal-related failures including thermal stress, cooling system malfunction, winding or core overheating, bushing aging, and tap changer damage.

5. The system of claim 1, wherein said sensor fusion engine correlates multivariate input comprising at least one of:acoustic signals with chemical markers to confirm partial discharges,temperature trends with chemical degradation to track insulation breakdown,acoustic signals with temperature anomalies to detect mechanical arcing or vibration- related heating, andelectrical signals with temperature to enable lifetime prediction and real-time risk / stress.

6. The system of claim 1, wherein the plurality of multiplexed optical fiber sensors are configured for detection of gas species selected from the group consisting of: hydrogen, acetylene, carbon monoxide, carbon dioxide, methane (and other relevant hydrocarbons), humidity, moisture, and furans.

7. The system of claim 1, wherein detection of all possible dissolved gases is enabled by multiple-channeled signal conditioning circuitry using galvanic isolation, with number of channels expandable through time-division multiplexing of isolated shift registers that reduce the number of input pins needed on the microcontroller.

8. The system of claim 1, wherein the plurality of multiplexed optical fiber sensors comprise acoustic sensors configured to operate over a frequency range of 1 kHz to 500 kHz or greater, and use multimode interferometers interrogated with VCSEL or high-power LED sources slightly above visible wavelengths.

9. The system of claim 1, wherein the plurality of multiplexed optical fiber sensors include outputs that are digitized by 16-bit analog-to-digital converter (ADC) with maximum allowed sampling rate of 2 MHz enable for high-fidelity capture of acoustic sensor signals for accurate real-time analysis of transformer operational anomalies.

10. The system of claim 1, wherein the plurality of multiplexed optical fiber sensors comprise temperature sensors comprising thermo-optic coatings and are collocated with chemical sensors to provide real-time compensation for temperature-induced signal variation in dissolved gas analysis.

11. The system of claim 1, further comprising a decision engine configured to:determine fault type based on classification confidence thresholds,calculate incipient failure probabilities, andgenerate real-time maintenance alerts and transformer life expectancy estimates.

12. The system of claim 1, further comprising additional sensors that do not employ fiber optic cables including current sensors, voltage sensors, hall-effect sensors, temperature sensors, humidity sensors, weather sensors, temperature sensors, tilt and vibration sensors.

13. A data analytics platform that houses the data from a plurality of transformers and processes it to incorporate relevant sensor and analytics outputs to inform conclusions, such as asset health monitoring and real-time operations optimization.

14. An edge processing based method that utilizes the system of claim 1.

15. The system of claim 1, wherein the system utilizes photonic integrated circuits for scalability, cost, manufacturability and / or performance advantages as compared to traditional integrated circuits and discrete optoelectronic components.

16. The system of claim 1, wherein the system enables energy independence of the signal interrogator by energy harvesting through an energy harvesting supercapacitor circuit and a splitcore current transformer that also act as an isolation transformer to provide medium voltage isolation.

17. The system of claim 1, wherein the sensor fusion engine is further configured to perform at least one of the following:perform signal processing schemes and their applications, including self referencing, denoising, and similar processes;integrate data into analytics platforms, such as established grid analytics including demand response and related;edge processing to store, process, and interpret data at the sensor node location; isolation of the itnerrogation system and related opto-electronics in terms of EMI and medium voltage operational environments, as well as environmental disturbances; andcombinations of fiber optic sensors with traditional sensors in the form of a multiparameter sensor system.

Images