{"record_key":"layered-fiber-optic-sensor-coating-and-edge-ready-ai-for-multi","identifier":"06905","external_key":"06905","title":"Layered Fiber-Optic Sensor Coating and Edge-Ready AI for Multi-Gas Sensing from a Single Sensor Point","summary":null,"url":"https://inventions.pitt.edu/technologies/layered-fiber-optic-sensor--06905","json_url":"https://inventions.pitt.edu/technologies/layered-fiber-optic-sensor--06905.json","object_key":"technologies","description":"This technology is an optical fiber sensor with a stack of concentric thin-film sorbent layers that lets one sensing point distinguish several analytes. It also includes a way to train compact, edge-deployable inference models that combine physics-simulated and real sensor data, enabling accurate multiparameter readouts without relying on a central cloud.\nDescription\nUnmet Need - Existing multiparameter fiber sensors need multiple sensors or heavy computation to separate overlapping signals, and this technology addresses that by building analyte discrimination into the sensor structure and a physics-informed model.\nThe sensor is built around an optical waveguide, such as an optical fiber, whose exposed core is surrounded by a multilayered heterostructure of concentric thin-film sorbent layers (materials that selectively take up target molecules). The layers are designed so that molecules are sieved and filtered as they move inward, different layers interact differently with a given analyte, and light of different wavelengths penetrates to different depths within the stack. Together these features let a single sensor point separate the signals of multiple analytes. Optionally, a layer of plasmonic metal nanoparticles (tiny metal particles that respond strongly to light) can sit on the fiber and act as both an optical sensitizer and a temperature reference. The stack can be built by layer-by-layer self-assembly, such as sequential dip-coating.\nThe companion analytics framework trains an inference model on a mix of physics-simulated synthetic data and real sensor data. The simulated data acts as a regularizer, which helps the model avoid overfitting to limited lab measurements. Features are chosen with principal component analysis to pick optimal wavelengths. A recurrent neural network (LSTM) with a drift-compensation cell handles long-term drift and transient effects. The trained model is stored on a compact, edge-compatible microcontroller so readouts are produced in real time near the sensor. A nonlinear Kalman filter variant estimates top-oil and hot-spot temperatures of a distribution transformer from a fiber-optic temperature sensor.\nApplications\n- Multi-gas monitoring for energy infrastructure and grid-tied IoT sensor networks\n- Condition monitoring of distribution transformers (top-oil and hot-spot temperature estimation)\n- Optical “electronic nose” or “photonic nose” systems for complex gas mixtures\n- Sensing in hazardous, explosive, or flammable environments where electrical sensors are impractical\n- Edge-based calibration and inference software for other optical and fiber sensing platforms\nAdvantages\n- Multiple analytes resolved from one compact sensor node, reducing the need for sensor arrays or bundles\n- Built-in molecular sieving and layer-specific interactions that improve analyte discrimination\n- Wavelength-dependent penetration depth that provides optical multiplexing within a single coating\n- Self-referenced temperature sensing through the plasmonic nanoparticle layer\n- Physics-informed training with drift compensation, deployable on low-cost edge microcontrollers\nInvention Readiness\nA prototype sensor has been built and tested, combining a ZIF-8 metal-organic framework sorbent layer with a gold overlayer. It showed two distinct spectral responses to CO₂ and temperature-dependent CO₂ and N₂ adsorption behavior. The analytics framework has been demonstrated on simulated single-sensor and two-sensor datasets, with median concentration prediction errors of 0.44% (single sensor) and 0.24% and 0.77% (two-sensor, gases 1 and 2).\nIP Status\nhttps://patents.google.com/patent/WO2026085174A1","attributes":[{"key":"file_number","label":"Reference Number","values":["06905"]},{"key":"technology_type","label":"Technology Type","values":["Engineering Technology"]},{"key":"technology_subtype","label":"Technology Subtype","values":["Sensor"]},{"key":"lead_inventor_name","label":"Lead Inventor","values":["Paul Ohodnicki"]},{"key":"lead_inventor_dept","label":"Department","values":["Mechanical Engineering and Materials Science"]},{"key":"all_tech_innovators","label":"All Innovators","values":["Michael Paul Buric","Nageswara Rao Lalam","Paul Richard Ohodnicki","Enrico Sarcinelli","Yang-Duan Su","Ruishu Wright","Jeffrey K. Wuenschell"]},{"key":"technology_readiness_level","label":"Technology Readiness Level","values":["3. Prototype development"]},{"key":"date_submitted","label":"Date Submitted","values":["Sep 27, 2024"]}],"related":[]}