System and method for differentiating reflections of fire or harmless flames

Abstract

The present disclosure relates to systems and methods for differentiating reflections of fire or harmless flames. A system is disclosed. The system includes a flame detector configured to detect radiation within a field of view (FOV) and convert the radiation into one or more analog-to-digital converter (ADC) signals. The at least one processor is operably coupled to the at least one flame detector. The at least one processor is configured to receive the one or more ADC signals from the at least one flame detector and determine a plurality of characteristics from the one or more ADC signals. Further, the plurality of characteristics includes at least one of a statistical characteristic, a frequency-based characteristic, or a time-based characteristic. The at least one processor is then configured to determine, using a trained machine learning (ML) model, whether the one or more ADC signals indicate a fire, an innocent flame, or a reflection of an innocent flame based at least on the plurality of characteristics.

Description

Technical Field

[0001] Example embodiments generally relate to a system and method, and more particularly to a flame detector for determining reflections of a fire or harmless flame. Background Art

[0002] Flame detectors are safety devices designed to identify flames generated by the combustion of different fuel sources and to promptly alert occupants or operators to the presence of fire and smoke. Existing flame detectors may not be able to distinguish between actual fires and harmless flames or reflections of harmless flames. For example, industrial plants such as refineries, chemical plants, and natural gas processing plants often utilize gas combustion equipment (such as flare stacks) to process gas. Flames often emerge from flare stacks, which is expected and known. Industrial plants often include structures with shiny surfaces, such as metal silos. "Harmless" flames emerging from flare stacks can cause reflections on the shiny surfaces, which may be sensed by flame detectors. Existing flame detectors often mistakenly trigger alarms for detecting reflections of harmless flames emerging from flare stacks.

[0003] Additionally, bright lights and / or high-intensity glare are common in many industrial processes and domestic environments. These bright lights and / or high-intensity glare can also cause existing flame detectors to falsely trigger alarms designed to detect these lights. Furthermore, certain events occurring on industrial sites, such as welding, can also cause false alarms to be triggered.

[0004] Applicants have identified many areas for improvement in existing technologies and methods that are the subject of the embodiments described herein. Through effort, ingenuity, and innovation, many of these deficiencies, challenges, and problems have been addressed by developing solutions included in the embodiments of the present disclosure, some examples of which are described in detail herein. Summary of the Invention

[0005] The following presents a brief overview of some example embodiments to provide a basic understanding of some aspects of the present disclosure. This summary is not an exhaustive overview and is not intended to identify key or important elements, nor to describe the scope of such elements. It should also be understood that the scope of the present disclosure encompasses many possible embodiments in addition to those summarized here, some of which will be further described in the detailed description presented below.

[0006] In one example embodiment, a system is disclosed. The system includes at least one flame detector configured to detect one or more radiations within a field of view (FOV) and convert the radiation into one or more analog-to-digital converter (ADC) signals. Furthermore, at least one processor is operably coupled to the at least one flame detector, the at least one processor configured to receive the one or more ADC signals from the at least one flame detector and determine a plurality of characteristics based on the one or more ADC signals. The plurality of characteristics include at least one of statistical characteristics, frequency-based characteristics, or time-based characteristics. Furthermore, the at least one processor is configured to use a trained machine learning (ML) model to determine, based at least on the plurality of characteristics, whether the one or more ADC signals indicate a fire, a harmless flame, or a reflection of a harmless flame.

[0007] In some embodiments, the at least one processor is configured to train the ML model based at least on the plurality of features extracted from the one or more ADC signals over a period of time. In some embodiments, the at least one processor is configured to deploy the trained ML model for determining the fire, the harmless flame, or a reflection of the harmless flame. In some embodiments, the at least one flame detector comprises at least one of an infrared (IR) sensor, a photodiode, or a combination of an IR sensor and a photodiode.

[0008] In some embodiments, the statistical characteristic comprises at least one of skewness, kurtosis, or a ratio of skewness to kurtosis. In some embodiments, the at least one processor is configured to detect fluctuations or modulations in the plurality of characteristics determined based on the one or more ADC signals to determine whether the one or more ADC signals are indicative of the fire, the harmless flame, or a reflection of the harmless flame.

[0009] In some embodiments, the at least one processor is configured to extract the plurality of characteristics of the flame and the reflection of the harmless flame in a low-frequency range and a high-frequency range. In some embodiments, the low-frequency range is defined as a range between 2 Hz-9 Hz and 11 Hz-15 Hz, and the high-frequency range is defined as frequencies above 15 Hz. In some embodiments, the trained ML model comprises an aggregated simulation of multiple models trained by the at least one processor over a period of time.

[0010] In some embodiments, the at least one processor is configured to generate a signal in response to determining that the one or more ADC signals are indicative of a fire, and transmit the signal to a communication device to alert a user.

[0011] In another example embodiment, a method is disclosed. The method includes receiving, via at least one processor, one or more analog-to-digital (ADC) signals from at least one flame detector. The method also includes determining, via the at least one processor, a plurality of characteristics based on the one or more ADC signals. The plurality of characteristics includes at least one of statistical characteristics, frequency-based characteristics, or time-based characteristics. The method also includes determining, via the at least one processor, using a trained machine learning (ML) model, whether the one or more ADC signals indicate a fire, a harmless flame, or a reflection of a harmless flame based at least on the plurality of characteristics.

[0012] The above summary of the invention is provided only for the purpose of summarizing some exemplary embodiments to provide a basic understanding of some aspects of the present disclosure. Therefore, it should be understood that the above embodiments are merely examples and should not be construed as narrowing the scope or essence of the present disclosure in any way. It should be understood that the scope of the present disclosure encompasses many possible embodiments in addition to those summarized here, some of which will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Having thus generally described certain example embodiments of the present disclosure, reference will now be made to the accompanying drawings, which are not necessarily drawn to scale, and in which:

[0014] Figure 1 shows a block diagram of a system according to an example embodiment of the present disclosure;

[0015] Figure 2 shows a perspective view of at least one flame detector according to an example embodiment of the present disclosure;

[0016] Figure 3 A flow chart illustrating a system according to an example embodiment of the present disclosure is shown;

[0017] Figure 4 shows a graphical representation of a spectrum of one or more radiations according to an example embodiment of the present disclosure;

[0018] Figure 5A shows a graphical representation of one or more analog-to-digital converter (ADC) signals and one or more ADC signals filtered in a bandpass frequency range according to an example embodiment of the present disclosure;

[0019] Figure 5B shows a graphical representation of one or more ADC signals and one or more ADC signals filtered at a low frequency and at a high frequency according to an example embodiment of the present disclosure;

[0020] Figures 6A to 6Fshows a graphical representation of a kernel density estimation (KDE) plot of a plurality of characteristics determined from one or more ADC signals according to an example embodiment of the present disclosure;

[0021] Figure 7 shows a graphical representation of a learning curve for a machine learning (ML) model according to an example embodiment of the present disclosure;

[0022] Figure 8 A graphical representation of a confusion matrix showing ML model testing results according to an example embodiment of the present disclosure is shown;

[0023] Figure 9 shows a receiver operating characteristic (ROC) plot of a trained ML model according to an example embodiment of the present disclosure;

[0024] Figure 10 shows a table showing test results of a trained ML model according to an example embodiment of the present disclosure; and

[0025] Figure 11 A flow chart illustrating a method according to an example embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0026] Some embodiments will now be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all, embodiments of the present disclosure are shown. Indeed, the various embodiments may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements.

[0027] The components illustrated in the drawings represent components that may or may not be present in the various embodiments of the present disclosure described herein, such that an embodiment may include fewer or more components than those shown in the drawings without departing from the scope of the present disclosure. Some components may be omitted from one or more of the drawings or shown in phantom to make underlying components visible.

[0028] As used herein, the term "comprising" means including but not limited to, and should be interpreted in the manner in which it is typically used in a patent context. The use of broader terms such as "including," "comprising," and "having" should be understood to provide support for narrower terms such as "consisting of," "consisting essentially of," and "composed essentially of."

[0029] The phrases "in various embodiments," "in one embodiment," "according to one embodiment," "in some embodiments," etc. generally mean that the particular feature, structure, or characteristic following the phrase may be included in at least one embodiment of the present disclosure, and may be included in more than one embodiment of the present disclosure (importantly, such phrases are not necessarily referring to the same embodiment).

[0030] The word “example” or “exemplary” is used herein to mean “serving as an example, instance, or illustration.” Any implementation described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other implementations.

[0031] If the specification states that a component or feature "may," "could," "might," "should," "will," "preferably," "likely," "typically," "optionally," "for example," "usually," or "might" (or other such language) be included or have a characteristic, that particular component or feature is not required to be included or have that characteristic. Such a component or feature may optionally be included in some embodiments, or it may be excluded.

[0032] Embodiments of the present disclosure will be described more fully below with reference to the accompanying drawings, in which like numerals represent like elements throughout the several figures, and in which example embodiments are shown. However, embodiments of the present disclosure can be embodied in alternative forms and should not be construed as limited to the embodiments set forth herein. The examples set forth herein are non-limiting examples and are merely examples of other possible examples.

[0033] The present invention provides various embodiments of systems and methods for distinguishing between fire, harmless flames, or reflections of harmless flames. Embodiments may be configured to use at least one flame detector to detect one or more radiations emitted within a field of view (FOV). Embodiments may be configured to convert the detected one or more radiations into one or more analog-to-digital converter (ADC) signals. Embodiments may be configured to determine multiple characteristics based on the one or more ADC signals. Based on the determined multiple characteristics, embodiments may be configured to use a trained machine learning (ML) model to determine whether the one or more ADC signals indicate a fire, a harmless flame, or a reflection of a harmless flame. Embodiments may also be configured to trigger an alarm to alert a user to the presence of a fire within the FOV. Embodiments may notify a user of the presence of a harmless flame reflection within the FOV. Embodiments may prevent any false alarms caused by the reflection of a harmless flame resembling a flame source, thereby preventing any unnecessary panic in the surrounding environment. Furthermore, embodiments may be integrated into various settings, from industrial facilities to domestic applications, providing an accurate solution for determining the presence of a fire, a harmless flame, or a reflection of a harmless flame to prevent any false alarms and panic in the surrounding environment.

[0034] Figure 1 A block diagram of a system 100 for determining a fire, a harmless flame, or a reflection of a harmless flame is shown according to an example embodiment of the present disclosure. Figure 2 A perspective view of at least one flame detector 104 is shown according to an example embodiment of the present disclosure.

[0035] System 100 may include at least one flame detector 104, at least one processor 106, memory 108, a machine learning (ML) model 110, input / output circuitry 112, a communication device 114, and communication circuitry 116. In some embodiments, at least one flame detector 104 may be configured to detect one or more radiations emitted from a flame source 102 within a field of view (FOV). In some embodiments, flame source 102 may correspond to a source of hydrogen, hydrocarbon gas, methane gas, or other combustible fuel. In some embodiments, flame source 102 may be configured to emit one or more radiations during combustion. Furthermore, the one or more radiations may be detected in the form of one or more infrared (IR) signals, one or more optical signals, or one or more heat signals.

[0036] In some embodiments, the at least one flame detector 104 can be configured to detect one or more types of radiation. In addition, the at least one flame detector 104 can include at least one subassembly 200 having a printed circuit board assembly (PCBA) stack 202. In some embodiments, the PCBA stack 202 can be made of at least one flame detector 104 having at least one infrared (IR) sensor 204, a photodiode 208, an ultraviolet (UV) light sensor 210, or a combination of at least one IR sensor 204, a photodiode 208, and a UV light sensor 210, such as Figure 2 As shown. In some embodiments, at least one flame detector 104 may correspond to at least one IR sensor 204 or photodiode 208. In some embodiments, at least one IR sensor 204 and photodiode 208 may be configured to capture one or more types of radiation. Additionally, UV light sensor 210 may be configured to capture one or more types of radiation in the ultraviolet frequency range. Furthermore, UV light sensor 210 may be coupled to UV test source 212. In some embodiments, UV test source 212 may be configured to receive a response from UV light sensor 210.

[0037] like Figure 2As shown, PCBA stack 202 may include at least one analog front-end (AFE) board. In some embodiments, the AFE board may be fabricated from at least one IR sensor 204. In some embodiments, the AFE board may be fabricated from a photodiode 208. In some embodiments, the AFE board may be fabricated from a combination of at least one IR sensor 204 and a photodiode 208. In some embodiments, at least one IR sensor 204 may also correspond to one or more lead selenide (PbSe) detectors. In some embodiments, the one or more PbSe detectors may include a broadband PbSe detector 206A and a long-band PbSe detector 206B. Furthermore, broadband PbSe detector 206A may be configured to detect one or more radiation within a broadband frequency band in the ultraviolet range. Furthermore, long-band PbSe detector 206B may be configured to detect one or more radiation within a long frequency band in the ultraviolet range. In some embodiments, at least one IR sensor 204 and photodiode 208 may be configured to capture one or more radiation in the form of one or more analog signals. Furthermore, the AFE board may be fabricated from one or more active and passive electronic components that may enable the at least one flame detector 104 to detect one or more radiations.

[0038] In some embodiments, at least one flame detector 104 can be integrated with an analog-to-digital converter (ADC) (not shown). In some embodiments, the ADC can be configured to convert one or more analog signals corresponding to one or more radiation into one or more digital signals. Furthermore, the one or more digital signals can also be referred to as one or more ADC signals, such as one or more ADC counts. In some embodiments, the ADC can be operably coupled to at least one processor 106. In some embodiments, the ADC can be configured to feed the one or more ADC signals to the at least one processor 106.

[0039] In some embodiments, the at least one processor 106 may include suitable logic components, circuits, and / or interfaces that are operable to execute one or more instructions stored in the memory 108 in order to perform predetermined operations. In one embodiment, the at least one processor 106 may be configured to decode and execute any instructions received from one or more other electronic devices or servers. The at least one processor 106 may be configured to execute one or more computer-readable program instructions, such as program instructions for performing any functions described in this specification. In addition, the at least one processor 106 may be implemented using one or more processor technologies known in the art. Examples of the at least one processor 106 include, but are not limited to, one or more general-purpose processors and / or one or more special-purpose processors (e.g., a digital signal processor or a field programmable gate array (FPGA) processor).

[0040] In some embodiments, at least one processor 106 may be configured to receive one or more ADC signals from the ADC. Furthermore, at least one processor 106 may be communicatively paired with ML model 110. In some embodiments, at least one processor 106 may be configured to implement one or more machine learning protocols to distinguish between a fire, a harmless flame, or a reflection of a harmless fire. In some embodiments, the one or more protocols may involve collecting a data set and extracting a plurality of features from the collected data set. Furthermore, at least one processor 106 may be configured to train ML model 110 based at least on the plurality of features extracted from the one or more ADC signals over a period of time. In some embodiments, at least one processor 106 determines that the one or more ADC signals indicate a fire, and then at least one processor 106 may transmit the signal to communication device 114 via input / output circuitry 112 and communication circuitry 116.

[0041] In some embodiments, ML model 110 may be configured to validate and test multiple characteristics by comparing them to multiple scenarios over a period of time. The ML model comprises an aggregated simulation of multiple models trained by at least one processor over a period of time. Furthermore, the multiple scenarios may correspond to instances of fires generated from different flame sources and reflections of harmless fires. Furthermore, the trained ML model 110 may be configured to generate one or more results based at least on the validation and testing of the multiple characteristics. In some embodiments, the one or more results may indicate a fire, a harmless flame, or a reflection of a harmless fire. Examples of harmless flames may include, but are not limited to, flare stacks. Furthermore, a flare stack may correspond to one or more situations in which a harmless flame is generated to eliminate one or more hazardous gases. Furthermore, the one or more hazardous gases may be generated in various industrial environments (i.e., refineries, chemical plants, gas processing plants, and at oil or gas production sites). In some embodiments, the results of the validation and testing may be stored in memory 108. In some embodiments, memory 108 may be configured to store a set of instructions and data for execution by one or more processors 106. The memory 108 may include one or more instructions that can be executed by at least one processor 106 to perform specific operations. It will be apparent to a skilled person that the one or more instructions stored in the memory 108 enable the hardware of the system 100 to perform the predetermined operations. Some well-known memory 108 implementations include, but are not limited to, fixed (hard) drives, magnetic tapes, floppy disks, optical disks, compact disk read-only memories (CD-ROMs) and magneto-optical disks, semiconductor memories (such as ROMs), random access memories (RAMs), programmable read-only memories (PROMs), erasable PROMs (EPROMs), electrically erasable PROMs (EEPROMs), flash memory, magnetic or optical cards, or other types of media / machine-readable media suitable for storing electronic instructions.

[0042] like Figure 1 As shown, system 100 may include input / output circuitry 112 that enables a user to communicate or interact with system 100 via communication device 114. Note that input / output circuitry 112 may serve as a medium for transmitting input from communication device 114 to or from system 100. In some embodiments, input / output circuitry 112 may refer to hardware and software components that facilitate the exchange of information between a user and system 100. Input / output circuitry 112 may include various input devices, such as a keyboard, a barcode scanner, a GUI for a user to provide data, and various output devices, such as a display, a printer, for a user to receive data.

[0043] In one example, the communication device 114 may include N user devices. In some embodiments, the communication device 114 may include a graphical user interface (GUI) (not shown) as an input circuit to allow a user to input data. In some embodiments, the communication device 114 may include at least one of one or more mobile phones, laptop computers, etc.

[0044] In some embodiments, the communication circuitry 116 can allow the system 100 and the communication device 114 to exchange data or information with other systems or devices. In addition, the system 100 can be communicatively coupled to a network interface via one or more protocols and software modules for sending and receiving data or information. In some embodiments, the communication circuitry 116 can include an Ethernet port, a Wi-Fi adapter, or a communication protocol for connecting to other systems. The communication circuitry 116 can allow the system 100 to stay up to date.

[0045] It will be apparent to those skilled in the art that the above-described components of the system 100 are provided for illustrative purposes only without departing from the scope of the present disclosure.

[0046] Figure 3 A flow chart 300 of the system 100 is shown according to an example embodiment of the present disclosure. Figure 4 Shown is a graphical representation of a spectrum 400 of one or more radiations according to an example embodiment of the present disclosure. Figure 5A Shown are one or more analog-to-digital converter (ADC) signals 502 and a graphical representation of the one or more ADC signals filtered in a bandpass frequency range, according to an example embodiment of the present disclosure. Figure 5B 1 shows a graphical representation of one or more ADC signals 502 and one or more ADC signals 502 filtered at a low frequency and one or more ADC signals filtered at a high frequency according to an example embodiment of the present disclosure. Figure 3 right Figures 4 to 5B Provide a description.

[0047] At operation 302, at least one flame detector 104 may be configured to detect one or more radiations from the flame source 102 and other infrared sources. The at least one flame detector 104 may include at least one IR sensor 204, a photodiode 208, a UV light sensor 210, and a combination of at least one IR sensor 204, a photodiode 208, and a UV light sensor 210. Furthermore, the at least one IR sensor 204 may detect one or more radiations in the range of 2.5 μm to 3.0 μm. Furthermore, the photodiode 208 may be configured to detect one or more radiations in the range of 0.18 μm to 0.26 μm. In some embodiments, the at least one IR sensor 204 and the photodiode 208 may be configured to detect water vapor generated by the combustion of the flame source 102, which may include, but is not limited to, hydrogen and hydrocarbon gases.

[0048] At operation 304, at least one flame detector 104 may be configured to detect one or more radiations in the form of electromagnetic signals. In some embodiments, the one or more electromagnetic signals may correspond to one or more electrical signals. In some embodiments, the one or more electromagnetic signals may have different wavelengths. A graphical representation of a spectrum 400 of one or more radiations emitted from flame source 102 may be provided at Figure 4 . The graphical representation may represent a spectrum 400 of one or more wavelengths of radiation emitted by flame source 102. In one exemplary embodiment, spectrum 400 may include one or more wavelengths of radiation emitted due to the combustion of ethylene (shown by 402). Additionally, spectrum 400 may include one or more wavelengths of radiation emitted due to sunlight (shown by 404). In another exemplary embodiment, spectrum 400 may include one or more wavelengths of radiation emitted due to the combustion of hydrogen (shown by 406).

[0049] In some embodiments, at least one flame detector 104 can be configured to detect one or more radiations in the form of analog signals. Furthermore, the one or more radiations can correspond to one or more electromagnetic radiations in the form of infrared (IR), visible light, and ultraviolet (UV) wavelengths. Furthermore, the wavelengths of the one or more radiations depend on the type of fuel source.

[0050] At operation 306, the analog-to-digital converter (ADC) 308 may be configured to acquire one or more analog signals and convert them into one or more ADC signals 502. Furthermore, the at least one processor 106 may be configured to receive the one or more ADC signals 502 from the ADC 308. Furthermore, the at least one processor 106 may be configured to determine a plurality of characteristics based on the one or more ADC signals 502. In some embodiments, the plurality of characteristics may include at least one of frequency-based characteristics, time-based characteristics, or statistical characteristics.

[0051] At operation 310, at least one processor 106 may be configured to pass one or more ADC signals 502 through at least one filter to remove any noise present in the one or more ADC signals 502. Furthermore, filtering the one or more ADC signals 502 may be configured to emphasize a signal-to-noise ratio. In some embodiments, the at least one filter may correspond to a bandpass filter, a high-pass filter, or a low-pass filter. Furthermore, after passing the one or more ADC signals 502 through the bandpass filter, the high-pass filter, or the low-pass filter, the one or more ADC signals are filtered.

[0052] The filtered one or more ADC signals may correspond to a narrowband alternating current (NB_AC) signal. Furthermore, after filtering the one or more ADC signals 502 with a narrowband, the NB_AC signal may be obtained. Furthermore, the filtered one or more ADC signals may correspond to a near-infrared direct current (NIR_DC) signal. Furthermore, after filtering the DC component of the one or more ADC signals 502 received from the near-infrared sensor, the NIR_DC signal may be obtained.

[0053] Furthermore, the filtered one or more ADC signals may correspond to a low-band alternating current (LB_AC) signal. Furthermore, after filtering the one or more ADC signals 502 in the low-band, the LB_AC signal may be obtained. Furthermore, the filtered one or more ADC signals may correspond to a low-band direct current (LB_DC) signal. Furthermore, after filtering the one or more ADC signals 502 in the low-band, the LB_DC signal may be obtained.

[0054] Furthermore, the filtered one or more ADC signals 502 may correspond to a wideband alternating current (WB_AC) signal. Furthermore, after filtering the one or more ADC signals 502 with a wideband, a WB_AC signal may be obtained. Furthermore, the filtered one or more ADC signals 502 may correspond to a wideband direct current (WB_DC) signal. Furthermore, after filtering the one or more ADC signals 502 with a wideband, a WB_DC signal may be obtained.

[0055] In some embodiments, at least one processor 106 can be configured to pass one or more ADC signals 502 through at least one filter to remove any noise present in the one or more ADC signals 502. In some embodiments, the one or more ADC signals 502 can include a bandpass signal 504, a low-frequency signal 506, and a high-frequency signal 508. Furthermore, the low-frequency signal 506 defines a range between 2 Hz-9 Hz and 11 Hz-15 Hz, and the high-frequency signal 508 defines a frequency above 15 Hz. Furthermore, the filtering of the one or more ADC signals 502 can be configured to emphasize the signal-to-noise ratio.

[0056] In some embodiments, one or more ADC signals 502 and one or more ADC signals filtered with a bandpass frequency range are Figure 5A In addition, one or more ADC signals 502 and one or more ADC signals 502 filtered at a low frequency, one or more ADC signals filtered at a high frequency are shown in FIG. Figure 5B Shown in.

[0057] At operation 312, at least one processor 106 may be configured to perform data preprocessing and exploratory data analysis (EDA) on the filtered one or more ADC signals. In some embodiments, at least one processor 106 may be configured to apply a signal processing technique to the one or more ADC signals 502. In addition, the signal processing technique may correspond to a fast Fourier transform (FFT). In some embodiments, at least one processor 106 may be configured to find frequency-based features and time-based features based on at least the data preprocessing and EDA.

[0058] At operation 314, the at least one processor 106 may be configured to determine a power spectral density in the one or more ADC signals 502 based at least on data preprocessing and exploratory data analysis (EDA). Furthermore, the power spectral density may correspond to a power distribution of the one or more ADC signals 502 at one or more frequency spectra. At operation 316, the at least one processor 106 may be configured to determine a dominant frequency and peak value in the power spectrum based on the at least one power spectral density. In some embodiments, the dominant frequency may correspond to a flame pattern such as flickering and modulation.

[0059] At operation 318, the at least one processor 106 may be configured to identify a quiet region and an ambient region from the one or more ADC signals 502 based at least on the dominant frequencies and peaks in the power spectrum. In some embodiments, the quiet region may correspond to a frequency range in the one or more ADC signals 502 that may be configured to indicate a normal state when no flame is present. Additionally, the ambient region may correspond to a frequency range in the one or more ADC signals 502 that may be configured to indicate the presence of a flame.

[0060] At operation 320, at least one processor 106 may be configured to identify a dominant signature and a transition signature based on at least the quiet region and the ambient region. In some embodiments, the dominant signature may correspond to a flame pattern detected by the one or more ADC signals 502. In some embodiments, the transition signature may correspond to a sudden change in the one or more ADC signals 502.

[0061] In one embodiment, at operation 322, when the at least one processor 106 determines a dominant signature in the one or more ADC signals 502, the at least one processor 106 may be configured to find the dominant frequency and determine a rolling window. Furthermore, the rolling window method may correspond to frequency analysis of the one or more ADC signals 502 at different time intervals.

[0062] In another embodiment, at operation 324, when the dominant signature is not determined by the at least one processor 106, the at least one processor 106 may be configured to generate a frequency-based feature based on at least one of the one or more ADC signals 502. At operation 326, the at least one processor 106 may be configured to generate a plurality of ratio signals based on the low-frequency signal of the one or more ADC signals 502 and the high-frequency signal of the one or more ADC signals 502. In an exemplary embodiment, the plurality of ratio signals may include PSum_LbLo_WbHi: a ratio of PSumlow (the sum of at least one power spectral density over 2 Hz-9 Hz and 11 Hz-15 Hz) of the LBAC to PSumHigh (the sum of at least one power spectral density over 21 Hz-29 Hz) of the WBAC; and PPratioWbLb: a ratio of the slew rate limited peak-to-peak with fast decay of the WBAC to the slew rate limited peak-to-peak with fast decay of the LBAC.

[0063] At operation 328, at least one processor 106 may be configured to generate a time-based feature based at least on the one or more ADC signals 502. In some embodiments, the time-based feature may include mean variance, skewness, and kurtosis. At operation 330, at least one processor 106 may be configured to calculate the skewness and kurtosis. In some embodiments, the skewness and kurtosis may be configured to provide a statistical feature of the one or more ADC signals 502.

[0064] At operation 332, the at least one processor 106 may be configured to train the ML model 110 based on at least the determined plurality of characteristics. Furthermore, after training the ML model 110, the at least one processor 106 may be configured to validate the training data provided to the ML model 110. Furthermore, the trained ML model 110 may be configured to test the plurality of characteristics based on at least samples of the one or more ADC signals 502 provided to the at least one processor 106 via a rolling window method. The rolling window method may correspond to sampling the one or more ADC signals 502 every 16.67 ms / 120 points.

[0065] In some embodiments, the trained ML model 110 may include an aggregated simulation of multiple models that may be trained by at least one processor 106 over a period of time. In addition, the multiple models include at least one of logistic regression, random forest, support vector machine, gradient boosting, k-nearest neighbor, decision tree, or neural network. In some embodiments, the at least one processor 106 may be configured to validate and test the training data through one or more KDE plots. In addition, the KDE plot may be configured to provide a visual representation of the probability of the training results under different frequency ranges and different power spectra.

[0066] Figures 6A to 6F 1 shows a graphical representation of a kernel density estimation (KDE) plot of a plurality of characteristics determined from one or more ADC signals 502 according to an example embodiment of the present disclosure. Figure 3 right Figures 6A to 6F Provide a description.

[0067] In some embodiments, at least one processor 106 can be configured to validate the one or more ADC signals 502 provided to the ML model 110. Furthermore, the KDE graph 600 can illustrate a fire state (i.e., determining the probability of the one or more ADC signals 502 indicating a fire, a harmless fire, or a reflection of a harmless fire) for a narrowband alternating current (NB_AC) signal of the one or more ADC signals 502. Furthermore, the KDE graph 600 can illustrate a visual representation of the difference between the one or more ADC signals 502 indicating a fire and the one or more ADC signals 502 indicating a reflection of a harmless fire for the narrowband alternating current (NB_AC) signal of the one or more ADC signals 502. Furthermore, the X-axis of the KDE graph 600 corresponds to a value of a time interval, and the Y-axis of the KDE graph 600 corresponds to a cumulative density of the one or more ADC signals 502.

[0068] like Figure 6B As shown in FIG, at least one processor 106 can be configured to validate the one or more ADC signals 502 provided to the ML model 110. Furthermore, a KDE graph 602 can illustrate a fire state (i.e., determining a probability that the one or more ADC signals 502 indicate a fire, a harmless fire, or a reflection of a harmless fire) for a low-band direct current (LB_DC) signal of the one or more ADC signals 502. Furthermore, the KDE graph 602 can illustrate a visual representation of the difference between the one or more ADC signals 502 indicating a fire and the one or more ADC signals 502 indicating a reflection of a harmless fire for the low-band direct current (LB_DC) signal of the one or more ADC signals 502. Furthermore, an X-axis of the KDE graph 602 corresponds to a value of a time interval, and a Y-axis of the KDE graph 602 corresponds to a cumulative density of the one or more ADC signals 502.

[0069] like Figure 6C As shown in FIG, KDE graph 604 may illustrate the sum of the power spectral densities across the narrowband AC (PSDRatioHiSum NB_AC) signals of the one or more ADC signals 502. Furthermore, KDE graph 604 may illustrate a visual representation of the difference between the one or more ADC signals 502 indicating a fire and the one or more ADC signals 502 indicating reflections of a harmless flame, for the sum of the power spectral densities across the narrowband AC signals of the one or more ADC signals 502. Furthermore, the X-axis of KDE graph 604 corresponds to the value of the time interval, and the Y-axis of KDE graph 604 corresponds to the cumulative density of the one or more ADC signals 502.

[0070] like Figure 6DAs shown in FIG, KDE graph 606 may illustrate another sum of the power spectral densities across the narrowband AC (PSDRatioHiSum2 NB_AC) signals of the one or more ADC signals 502. Furthermore, KDE graph 606 may illustrate a visual representation of the difference between the one or more ADC signals 502 indicating a fire and the one or more ADC signals 502 indicating reflections of a harmless flame, for the sum of the power spectral densities across the narrowband AC signals of the one or more ADC signals 502. Furthermore, the X-axis of KDE graph 606 corresponds to the value of the time interval, and the Y-axis of the KDE graph corresponds to the cumulative density of the one or more ADC signals 502.

[0071] like Figure 6E As shown in FIG, KDE graph 608 may illustrate another sum of the power spectral densities over the narrowband AC signals at low frequencies (PSDRatioLowSum NB_AC) of the one or more ADC signals 502. Furthermore, KDE graph 608 may illustrate a visual representation of the difference between the one or more ADC signals 502 indicating a fire and the one or more ADC signals 502 indicating reflections of a harmless flame, for the sum of the power spectral densities over the narrowband AC signals of the one or more ADC signals 502. Furthermore, the X-axis of KDE graph 608 corresponds to the value of the time interval, and the Y-axis of KDE graph 608 corresponds to the cumulative density of the one or more ADC signals 502.

[0072] like Figure 6F As shown, KDE graph 610 shows a visual representation of the difference between one or more ADC signals 502 indicating a fire and one or more ADC signals 502 indicating reflections of a harmless flame, for the sum of the power spectral densities over the ratio of the narrowband signal and the lowband signal of the one or more ADC signals 502. Furthermore, the X-axis of KDE graph 610 corresponds to the value of the time interval, and the Y-axis of KDE graph 610 corresponds to the cumulative density of the one or more ADC signals 502.

[0073] Figure 7 700 of a learning curve for a machine learning (ML) model according to an example embodiment of the present disclosure. Figure 3 right Figure 7 Provide a description.

[0074] In some embodiments, the ML model 110 can be configured to record the loss of one or more ADC signals 502 during training through multiple features and validation of the trained ML model 110. In some embodiments, the X-axis shows the training set size of the dataset used to train the ML model 110. Additionally, the Y-axis shows the loss of the training set.

[0075] Figure 8 A graphical representation of a confusion matrix 800 illustrating ML model testing results according to one or more embodiments of the present disclosure. Figure 3 right Figure 8 Provide a description.

[0076] In some embodiments, the ML model 110 can employ a confusion matrix 800 to test a data set provided to the ML model 110 during training. In some embodiments, the confusion matrix 800 can be configured to provide predictions of test results. In some embodiments, a first quadrant 802 of the confusion matrix 800 provides predictions that are true positives, a second quadrant 804 of the confusion matrix 800 provides predictions that are true negatives, a third quadrant 806 of the confusion matrix 800 provides predictions that are false positives, and a fourth quadrant 808 of the confusion matrix 800 provides predictions that are false negatives.

[0077] Figure 9 Receiver Operating Characteristic (ROC) graph 900 of the trained ML model 110 according to one or more embodiments of the present disclosure is shown. Figure 3 right Figure 9 Provide a description.

[0078] In some embodiments, the ROC property of the ML model 110 can be configured to plot the true positive rate versus the false positive rate at various classification thresholds. The ROC property can be configured to provide a comprehensive visualization of the accuracy of the ML model 110. The X-axis of the ROC represents the false positive rate, and the Y-axis of the ROC represents the true positive rate.

[0079] At operation 334, the ML model 110 may be configured to determine the status of the flame by employing a voting classifier. Furthermore, the voting classifier may be configured to determine the status of the flame based at least on the results of the test performed by the ML model 110. Furthermore, at operation 346, the at least one processor 106 may be configured to generate a signal in response to determining that the one or more ADC signals 502 indicate a fire, a harmless flame, or a reflection of a harmless flame. Furthermore, the at least one processor 106 may be configured to transmit the signal to the communication device 114 for alerting a user.

[0080] In some embodiments, the trained ML model 110 may be configured to provide one or more test results based on tests performed by the ML model 110. Additionally, the test results relate to the response of the ML model 110 during combustion of the fuel source.

[0081] Figure 10 Shown is a table 1000 with test results of a trained ML model 110 , according to an example embodiment of the present disclosure.

[0082] In some embodiments, the test results of the ML model 110 are depicted in table 1000. Table 1000 provides a fuel or pseudo-fuel stimulus 1002, a condition 1004, and a predicted accuracy 1006. The fuel or pseudo-fuel stimulus 1002 provides a first reading of 0.1% heptane, which is associated with the flame detection condition. Detection of a flame carrying heptane fuel is predicted with 100% accuracy. The fuel or pseudo-fuel stimulus 1002 provides a second reading of 0.2% IPA, which is associated with the flame detection condition. Detection of a flame carrying IPA (isopropyl alcohol) fuel is predicted with 100% accuracy. The fuel or pseudo-fuel stimulus 1002 provides a third reading of 0.8% methane, which is associated with the flame detection condition. Detection of a flame carrying methane fuel is predicted with 100% accuracy. The fuel or pseudo-fuel stimulus 1002 provides a fourth reading of a harmless fire, which is associated with the flame detection condition. Detection of a harmless fire is predicted with 100% accuracy. Results for determining flame states for different fuel sources. In some embodiments, the ML model 110 may have 100% accuracy in determining the state of the flame.

[0083] Figure 11 1 shows a flow chart of a method 1100 according to an example embodiment of the present disclosure. Figures 1 to 10 right Figure 11 Provide a description.

[0084] At operation 1102, one or more ADC signals 502 may be received by at least one processor 106 from at least one flame detector 104. Furthermore, one or more radiations within the FOV may be detected by the at least one flame detector 104. In some embodiments, the at least one flame detector 104 may correspond to at least one IR sensor 204, a photodiode 208, or a combination of at least one IR sensor 204 and a photodiode 208. Furthermore, the one or more radiations may correspond to one or more analog signals. Furthermore, the at least one flame detector 104 may be configured to transmit the one or more analog signals to an ADC. In some embodiments, the ADC may be configured to generate the one or more ADC signals 502. For example, the at least one flame detector 104 may be installed in an industrial environment. The at least one flame detector 104 detects one or more radiations from the combustion of the flame source 102. Furthermore, the at least one flame detector 104 transmits one or more electrical signals corresponding to the one or more radiations to the at least one processor 106.

[0085] At operation 1104, frequency-based features and time-based features from the ADC signal 502 may be determined by the at least one processor 106. Furthermore, the frequency-based features and time-based features may be referred to as a plurality of features, as explained in operations 312-328. For example, the at least one processor 106 may be configured to determine a plurality of features based on one or more ADC signals 502. Furthermore, the plurality of features may include time-based features and frequency-based features. Frequency-based features may include frequency power spectra, dominant frequencies, and the like.

[0086] At operation 1106, one or more ratios may be calculated based on at least the determined frequency-based features and the time-based features by the at least one processor 106. Furthermore, the one or more ratios may correspond to frequencies of the one or more ADC signals 502 at different time intervals, as explained in operations 324 to 330. For example, the at least one processor 106 is configured to calculate the ratios based on the determined frequency-based features and the time-based features.

[0087] At operation 1108, at least one processor 106 is configured to determine statistical characteristics of the calculated one or more ratios. In some embodiments, the statistical characteristics may include skewness or kurtosis and the skewness-to-kurtosis ratio, as explained in operation 330. For example, at least one processor 106 is configured to determine the statistical characteristics by using the calculated one or more ratios. In addition, the at least one processor may employ one or more data pre-processing techniques to determine the statistical characteristics.

[0088] At operation 1110, at least one processor 106 may be configured to train an ML model 110 based on at least a plurality of features. In some embodiments, the at least one processor 106 may be configured to train the ML model 110 by feeding the plurality of features at equal time intervals. In addition, the plurality of features may include fluctuations and modulations of one or more ADC signals 502. For example, the at least one processor 106 is configured to feed the determined statistical features into the machine learning (ML) model 110. In addition, the at least one processor 106 is configured to train the ML model 110 based on one or more data sets fed to the ML model 110 at different time intervals. In addition, the at least one processor 106 is configured to validate and test the data sets.

[0089] At operation 1112, at least one processor 106 may be configured to use the trained ML model 110 to determine whether one or more ADC signals 502 indicate a fire, a harmless flame, or a reflection of a harmless flame. In some embodiments, at least one processor 106 may be configured to train, validate, and test the ML model 110 based on at least a plurality of characteristics to determine the state of the flame. For example, at least one processor 106 trains, validates, and tests the ML model 110 based on at least a plurality of features. Furthermore, at least one processor 106 determines the state of the flame as a harmless flame based on the test results of the ML model 110. For example, at least one processor 106 is configured to determine, based on at least the test results of the ML model 110, whether one or more ADC signals 502 indicate a fire, a harmless flame, or a reflection of a harmless flame.

[0090] Embodiments may be configured to determine a state of a flame in a field of view (FOV). Embodiments may be configured to detect one or more radiations emitted from a flame source via at least one flame detector 104. Embodiments may be configured to indicate a fire, a harmless flame, or a reflection of a harmless flame in the FOV. Embodiments may be configured to determine a plurality of characteristics based on the one or more radiations via at least one processor 106. Embodiments may be configured to use an ML model 110 to determine whether the detected one or more radiations are indicative of a fire, a harmless flame, or a reflection of a harmless flame.

[0091] Those skilled in the art to which the present invention pertains will appreciate many modifications and other embodiments of the present invention as set forth herein, having benefited from the teachings presented in the foregoing description and the associated drawings. It will be appreciated, therefore, that the present invention is not limited to the specific embodiments disclosed, and that modifications and other embodiments are intended to be included within the scope of the appended claims. In addition, although the foregoing description and the associated drawings have described example embodiments in the context of certain example combinations of elements and / or functions, it will be understood that different combinations of elements and / or functions may be provided by alternative embodiments without departing from the scope of the appended claims. In this regard, for example, as may be set forth in some of the appended claims, it is also contemplated that combinations of elements and / or functions different from those explicitly described above are also contemplated. Although specific terms are employed herein, they are used only in a general and descriptive sense, and not for restrictive purposes.

Claims

1. A system, comprising: at least one flame detector configured to detect one or more radiations within a field of view (FOV) and convert into one or more analog-to-digital conversion (ADC) signals; and at least one processor operatively coupled to the at least one flame detector, the at least one processor configured to: receiving the one or more ADC signals from the at least one flame detector; determining a plurality of characteristics based on the one or more ADC signals, wherein the plurality of characteristics comprises at least one of statistical characteristics, frequency-based characteristics, or time-based characteristics; as well as A trained machine learning (ML) model is used to determine whether the one or more ADC signals indicate a fire, a harmless flame, or a reflection of a harmless flame based at least on the plurality of characteristics.

2. The system of claim 1 , wherein the at least one processor is configured to train the ML model based at least on the plurality of features extracted from the one or more ADC signals over a period of time.

3. The system of claim 2, wherein the at least one processor is configured to deploy the trained ML model for determining the fire, the harmless flame, or the reflection of the harmless flame. 4 . The system of claim 1 , wherein the at least one flame detector comprises at least one of an infrared (IR) sensor, a photodiode, or a combination of the IR sensor and the photodiode.

5. The system of claim 1, wherein the statistical feature comprises at least one of skewness, kurtosis, or a ratio of skewness to kurtosis.

6. The system of claim 1 , wherein the at least one processor is configured to detect fluctuations or modulations in the plurality of characteristics determined from the one or more ADC signals to determine whether the one or more ADC signals are indicative of the fire, the harmless flame, or the reflection of the harmless flame. 7 . The system of claim 1 , wherein the at least one processor is configured to extract the plurality of characteristics of the fire, the harmless flame, and the reflection of the harmless flame in a low frequency range and a high frequency range.

8. The system of claim 7, wherein the low frequency range is defined as a range between 2 Hz - 9 Hz and 11 Hz - 15 Hz, and the high frequency range is defined as frequencies above 15 Hz.

9. The system of claim 1, wherein the trained ML model comprises an aggregated simulation of multiple models trained by the at least one processor over a period of time.

10. The system of claim 1 , wherein the at least one processor is configured to: generating a signal in response to determining that the one or more ADC signals are indicative of a fire; and The signal is transmitted to a communication device to alert a user.

Images