Flame analytics system

Patent metadata

Field	Value
Publication number	US-12467692-B2
Application number	US-202318397833-A
Country	US
Kind code	B2
Filing date	Dec 27, 2023
Priority date	Nov 2, 2018
Publication date	Nov 11, 2025
Grant date	Nov 11, 2025

Abstract

A flame analytics system that may incorporate a burner, one or more sensors at the burner, a historical database connected to the one or more sensors, a model training module connected to the historical database, and a runtime algorithm module connected to the one or more sensors and the model training module. The runtime algorithm may compare realtime data from the one or more sensors and historical data from the model training module in accordance with a machine learning algorithm. The system may further incorporate a fault detection module connected to the runtime algorithm module, a fault diagnostics module connected to the fault detection module, and an enunciator connected to the fault detection module. The one or more sensors may also include having video or acoustic sensitivity of combustion in the burner.

First claim

What is claimed is:

1 . A flame analytics system, comprising:
a burner; one or more sensors at or about the burner, wherein the one or more sensors include two or more video sensors positioned at different angles with respect to a flame in the burner, and wherein the two or more video sensors are configured to capture video data pertaining to the flame, wherein the video data comprise 2D views of the flame; a processor configured to:
reconstruct the 2D views of the flame into 3D views of the flame, wherein the 3D views of the flame are utilized to distinguish the flame from a background of the flame in the video data, and
determine one or more parameters associated with the flame based on the 3D views of the flame;
a historical database connected to the one or more sensors; a model training module connected to the historical database, wherein a machine learning algorithm is trained based on the historical database; and a runtime algorithm module connected to the one or more sensors and the model training module, wherein the runtime algorithm module compares real time data from the one or more sensors and the historical data from the model training module in accordance with the machine learning algorithm.

2 . The flame analytics system of claim 1 , further comprising:
a fault detection module connected to the runtime algorithm module; and a fault diagnostics module connected to the fault detection module; and an enunciator connected to the fault detection module, wherein a burner fault is detected based on the comparison between the real time data and the historical data.

3 . The flame analytics system of claim 2 , wherein:
the historical database and the model training module operate offline; and the runtime algorithm module, the fault detection module, the fault diagnostics module and the enunciator operate online.

4 . The flame analytics system of claim 1 , wherein the one or more sensors comprise a microphone configured to sense sound generated by the flame.

5 . The flame analytics system of claim 1 , wherein the burner is associated with an industrial furnace.

6 . The flame analytics system of claim 1 ,
wherein the one or more parameters associated with the flame comprise one or more of steam pressure, fuel rate, flame strength, control state, O 2 level, firing rate, lockout status, limit status, stack temperature, and flame state, wherein the one or more sensors at or about the burner are selected for detecting one or more of a presence of the flame, an absence of the flame, a quantitative measure of a strength of the flame, first out, the control state, the firing rate, the steam pressure, the O 2 level, the stack temperature, the flame state, the flame strength, the lockout status, the limit status, fuel consumption, interlock status, rate limit status, rate source, cycle count, and run time.

7 . The flame analytics system of claim 1 , wherein the 3D views distinguish the flame from other flames in a multi-burner appliance, and wherein the one or more parameters determined from the 3D views of the flame comprise a flame size, characterizing flame shape, hot spots, oscillations, fluctuations or flicker, and a fundamental frequency of the flame.

8 . The flame analytics system of claim 1 , wherein flame video analytics are combined with an audio signal from a microphone located to sense sound generated by the flame.

9 . The flame analytics system of claim 1 , wherein data distinguished by the runtime algorithm module comprises one or more of flame shape, flame size, hot spots, spectra for CO 2 , water vapor, UV, IR, visible light flame fluctuations, operational state, firing rate, temperature, pressure, flow rate, value percentage opening, O 2 , stability of the flame, burner nozzle condition, flame-out safeguard, and flame categorization.

10 . The flame analytics system of claim 9 , wherein flame video analytics are combined with data corresponding to one or more of the flame shape, the flame size, the hot spots, the spectra for CO 2 , the water vapor, the UV, the IR, the visible light flame fluctuations, the operational state, the firing rate, the temperature, the pressure, the flow rate, the value percentage opening, O 2 , the stability of the flame, the burner nozzle condition, the flame-out safeguard, and the flame categorization.

11 . A method, for detecting a burner fault, comprising:
obtaining runtime data of a burner from combustion sensors associated with the burner; processing the runtime data by a machine learning mechanism, wherein the machine learning mechanism receives historical data of the burner from a database, and the runtime data, wherein the machine learning mechanism comprises a machine learning model, the machine learning model comprises an air side submodel; a fuel side submodel, and a combustion submodel, wherein the combustion submodel further incorporates video data provided by one or more video sensors of a combustion process of the burner, wherein the video data are post-processed before being sent to the machine learning model, wherein an image of the post-processed video is automatically segmented in that the image is binarized using a combination of pixel values in each color channel, and a resulting binary image being divided into adjacent geometric regions, wherein a statistical measure between the pixel values in each color channel is created for each segment of the image, and wherein statistical measures are input to the combustion submodel of the machine learning model; and comparing the historical data and the runtime data of the burner to detect the burner fault.

12 . The method of claim 11 , further comprising:
predicting functionality of the burner based on the detected burner fault.

13 . The method of claim 11 , wherein:
the air side submodel incorporates at least air flows and air pressures; the fuel side submodel incorporates at least fuel flows and fuel pressures; and the combustion submodel incorporates at least a ratio of air and fuel flows.

14 . The method of claim 11 , wherein the combustion submodel further incorporates acoustic data provided by one or more sound sensors detecting sounds of a combustion process in the burner, and wherein:
the acoustic data is post-processed before being sent to the machine learning model; a raw acoustic signal of acoustic data is measured; a frequency space transform is applied to the raw acoustic signal; a spectral centroid is extracted from the frequency space transform; the spectral centroid represents a center-of-mass of a frequency spectrum; and the center-of-mass includes information of the combustion process.

15 . A method, comprising:
reading sensor data from one or more sensors of a combustion system; reading video data capturing a flame of the combustion system from one or more video sensors, wherein the video data comprises at least 2D views of the flame, wherein the 2D views of the flame are reconstructed into 3D views of the flame to distinguish the flame from a background of the flame in the video data, and wherein one or more parameters associated with the flame are determined based on the 3D views of the flame; sending the sensor data and the video data to a data acquisition system for machine learning processing, wherein the video data comprises at least the one or more parameters associated with the flame; obtaining results from the data acquisition system; and tuning the combustion system based on the results.

16 . The method of claim 15 , wherein the combustion system is dynamically tuned for an optimal air-fuel ra

Assignees

• Honeywell Int Inc

Inventors

• Kucera David
• Taylor Curtis
• Mcdonald Jonathan
• Stewart Gregory
• Culler Wyatt

Classifications

• G06F16/219

  Managing data history or versioning (querying versioned data G06F16/2474; querying temporal data G06F16/2477) · CPC title

• F23N2241/04

  Heating water · CPC title

• F23N2229/00

  Flame sensors · CPC title

• F23N2223/36

  PID signal processing · CPC title

• F23N2223/48

  Learning / Adaptive control · CPC title