A system and method for visualizing alkali metals based on flame emission spectroscopy

Abstract

The application discloses an alkali metal visual detection system and method based on flame emission spectrum. The system comprises a flame atomic emission system, a multispectral imaging system and a processing unit. The core of the system is that the multispectral imaging system adopts a single imaging sensor and is provided with a multi-channel optical assembly. Each channel is aimed at a specific alkali metal and innovatively integrates a band-pass filter and an intensity filter with a specific transmittance. The design can project various alkali metals (such as sodium, potassium, lithium and cesium) with great differences in spectral radiation intensity into multiple spatially separated images in a single exposure. The processing unit directly establishes a nonlinear mapping relationship between image data and alkali metal concentration by using a pre-trained machine learning model, thereby realizing direct prediction of the concentration. The application solves the technical problems that the traditional method is difficult to synchronously detect strong and weak signals and needs complex calibration due to the "self-absorption effect".

Description

Technical Field

[0001] This invention relates to a visual detection system and method for alkali metals based on flame emission spectroscopy, belonging to the field of multispectral technology. Background Technology

[0002] Biomass energy, as a widely available and environmentally friendly renewable energy source, is playing an increasingly important role in the global energy structure. Through thermochemical conversion technologies such as combustion and gasification, biomass can be efficiently converted into electricity and heat. However, biomass fuels are generally rich in alkali metals such as potassium (K) and sodium (Na). During high-temperature combustion, these alkali metals are highly volatile and easily released, leading to severe corrosion, ash accumulation, and slagging on the heating surfaces of boilers and other equipment. This not only significantly reduces energy conversion efficiency but also increases the safety risks of system operation. Therefore, real-time and accurate online monitoring of the release behavior of alkali metals during combustion is crucial for optimizing combustion processes and ensuring the safe operation of equipment.

[0003] Currently, methods for detecting alkali metals during combustion are mainly divided into two categories: offline analysis and online monitoring. Offline analysis techniques, such as scanning electron microscopy (SEM), X-ray fluorescence spectroscopy (XRF), and X-ray diffraction (XRD), typically perform physical and chemical characterization on collected ash or sediments after combustion. Although these methods can determine the final form and content of alkali metals, they are essentially "post-analysis" and cannot provide time-resolved data on the dynamic release of alkali metals during combustion, thus failing to meet the needs of process control.

[0004] To achieve real-time monitoring, various online technologies have emerged. Among them, active measurement technologies, such as laser-induced breakdown spectroscopy (LIBS) and planar laser-induced fluorescence (PLIF), introduce an external laser source to excite target atoms, achieving high temporal and spatial resolution. However, these technologies typically require complex and expensive optical equipment, have extremely high requirements for optical path collimation, and are large systems, limiting their portability and making them difficult to widely apply in variable industrial environments.

[0005] In contrast, passive measurement techniques based on atomic self-emission spectroscopy, particularly flame atomic emission spectrometry (FES), have been widely studied due to their relatively simple systems and lower costs. Existing research has utilized FES technology for online measurement of gaseous alkali metal concentrations in waste incinerators and biomass burners. However, existing FES technologies still face the following pressing technical challenges:

[0006] Calibration is complex and accuracy is limited: Due to the existence of the "self-absorption effect," the spectral radiance of alkali metals is not a simple linear relationship with their actual concentration, especially in the high-concentration region where the nonlinear effect is more significant. To ensure measurement accuracy, traditional FES systems must rely on complex multi-point calibration procedures to correct for errors caused by this effect. This process is time-consuming and labor-intensive, and it is difficult to adapt to dynamically changing combustion conditions.

[0007] Simultaneous detection of multiple elements is challenging: Real combustion products often contain multiple alkali metals, and their characteristic spectral radiation intensities differ significantly (for example, sodium's spectral intensity is much higher than cesium's). Traditional single-channel or spectrometer scanning methods struggle to accurately image and quantify these elements with vastly different signal strengths in a single exposure, easily leading to overexposure of strong signals or submersion of weak signals.

[0008] Poor sample versatility: Existing FES detection systems are typically designed for samples in specific states (such as gases or specific solutions). In practical applications, the samples to be tested may be in various states, such as liquids, solid powders, or combustion fumes. Currently, there is a lack of a universal online alkali metal detection platform on the market that can flexibly adapt to samples in different states.

[0009] Therefore, there is an urgent need to develop a new type of detection system that can simultaneously perform real-time, synchronous, and visual detection of multiple alkali metals, adapt to various sample forms such as liquid, solid, and smoke, and fundamentally overcome the calibration complexity and accuracy limitations caused by nonlinear problems such as self-absorption effect, thereby achieving simple, fast, and accurate online measurement. Summary of the Invention

[0010] The technical problem to be solved by the present invention is to provide a visual detection system and method for alkali metals based on flame emission spectroscopy, so as to overcome the shortcomings of the prior art.

[0011] The technical solution of this invention is:

[0012] One aspect provides a visual detection system for alkali metals based on flame emission spectroscopy, the system comprising:

[0013] A flame atomic emission system configured to thermally excite multiple alkali metals in a sample to emit spectral radiation, the flame atomic emission system further comprising a sample introduction device for introducing the sample into the excitation region;

[0014] A multispectral imaging system configured to capture the spectral radiation, the multispectral imaging system comprising:

[0015] A single imaging sensor; and

[0016] An optical component includes multiple optical channels configured to simultaneously project the spectral radiation as multiple spatially separated images onto different regions of a single imaging sensor in a single exposure, and each optical channel includes:

[0017] A bandpass filter is used to separate the characteristic wavelength of a specific alkali metal; and

[0018] An intensity filter with a preset transmittance is used to equalize the spectral radiation intensity of different alkali metals;

[0019] A processing unit configured for:

[0020] Image data containing multiple spatially separated images is received from the single imaging sensor; and based on the preset correspondence between the multiple optical channels and the multiple spatially separated images, a pre-trained machine learning model is used to analyze each spatially separated image, wherein the machine learning model is used to establish a nonlinear mapping relationship between the image data and at least one feature of the multiple alkali metals, so as to achieve direct prediction of the feature.

[0021] Furthermore, the sample is in liquid form, and the sample introduction device includes an atomizer for converting the liquid sample into an aerosol.

[0022] Furthermore, the sample is in the form of a solid powder, and the sample introduction device includes a screw feeder and a pneumatic conveying device.

[0023] Furthermore, the sample is in the form of smoke, and the sample introduction device includes a combustion chamber for generating the smoke and a vacuum pump for delivering the smoke.

[0024] Furthermore, the plurality of alkali metals includes at least two selected from sodium, potassium, lithium and cesium.

[0025] Preferably, when the specific alkali metal is sodium, lithium, potassium and cesium, the transmittance of the intensity filter corresponding to the alkali metal is approximately 30%, 25%, 40% and 90%, respectively.

[0026] Furthermore, the pre-trained machine learning model is a convolutional neural network.

[0027] On the other hand, a visual detection method for alkali metals based on flame emission spectroscopy is provided, characterized by the following steps:

[0028] Provide a flame for thermally activating alkali metals;

[0029] The flow field dynamics characteristics of the flame are visualized and monitored using schlieren imaging technology;

[0030] Based on the visualized monitoring results of the flow field dynamics, the combustion parameters of the flame are adjusted to obtain the flame flow field structure; and

[0031] The sample to be tested is introduced into the flame flow field structure for flame emission spectroscopy detection.

[0032] The beneficial effects of this invention are: compared with the prior art,

[0033] 1) This invention employs a single imaging sensor combined with an optical assembly containing multiple optical channels. By simultaneously projecting multiple spatially separated images in a single exposure, the system can capture the dynamic release behavior of various alkali metals (such as sodium, potassium, lithium, and cesium) during combustion in real time. Specifically, each optical channel innovatively incorporates a combination of a bandpass filter and an intensity filter. The intensity filter is calibrated for transmittance based on the significant differences in spectral radiation intensity among different alkali metals (e.g., 30% for sodium and 90% for cesium). This fundamentally solves the technical challenge of accurately imaging simultaneously due to significant differences in signal intensity (e.g., the strong radiation of sodium versus the weak radiation of cesium) in traditional methods. It avoids the problems of overexposure of strong signals and submersion of weak signals, achieving synchronous, dynamic, and visual detection of multiple alkali metals.

[0034] 2) This invention utilizes a pre-trained machine learning model (such as a convolutional neural network) to analyze the acquired multispectral images. This model can establish a nonlinear mapping relationship between image data and alkali metal concentration, thereby achieving direct concentration prediction. This end-to-end prediction method effectively overcomes the nonlinearity problem between concentration and light intensity caused by the "self-absorption effect" in traditional flame emission spectroscopy, thus eliminating the cumbersome, time-consuming, and error-prone multi-point calibration procedure required in existing technologies. This not only significantly improves detection efficiency but also enhances the system's adaptability to dynamically changing operating conditions, greatly simplifies the measurement process, and significantly improves detection accuracy.

[0035] 3) By configuring a sample introduction device adaptable to different physical states, the core detection system of this invention can be compatible with processing samples in various forms, such as liquid solutions, solid powders, and combustion fumes. This enables its widespread application in various complex industrial and experimental scenarios, such as biomass combustion and waste incineration, overcoming the limitations of existing detection equipment that is typically only applicable to specific sample forms and has poor versatility, thus possessing high versatility and flexibility.

[0036] 4) The present invention adopts a single camera and integrated optical components (which can be manufactured through processes such as 3D printing). Compared with the scheme of using multiple cameras or large spectrometers, and active measurement technologies such as laser-induced breakdown spectroscopy (LIBS) that rely on expensive lasers, the hardware cost, size and complexity of the present invention are significantly reduced, which makes it possible to widely deploy and apply in variable industrial environments. The system has a compact structure, low cost and is easy to carry. Attached Figure Description

[0037] Figure 1 The overall functional block diagram of the alkali metal visualization detection system provided in this embodiment of the invention;

[0038] Figure 2 : A schematic diagram of the system structure for detecting liquid samples in this embodiment of the invention;

[0039] Figure 3 : A schematic diagram of the system structure for detecting solid powder samples in this embodiment of the invention;

[0040] Figure 4 : A schematic diagram of the system structure for detecting smoke morphology samples in an embodiment of the present invention;

[0041] Figure 5 : A schematic diagram of the specific structure of the multispectral imaging system in this embodiment of the invention.

[0042] List of reference numerals in the attached diagram:

[0043] 1: Detection system; 100: Flame atomic emission system; 101: Burner; 102: Atomizer; 103: Atomization chamber; 104: Baffle; 105: Screw feeder; 106: Microcontroller; 107: Vacuum generator; 108: Smoke generation chamber; 109: Smoke delivery pump; 200: Multispectral imaging system; 201: Imaging sensor; 202: Optical components; 203: Main structure; 204: Optical channel; 205: Bandpass filter; 206: Intensity filter; 300: Processing unit; 400: Schlieren imaging system; 401: Point light source; 402: Concave mirror; 403: Knife edge; 404: High-speed camera. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] The core inventive concept of this invention lies in providing a solution that deeply couples hardware and algorithms to overcome the nonlinear problem of the "self-absorption effect" in existing flame emission spectroscopy (FES) technology. Specifically, this invention first uses a specially designed multi-channel, intensity-equalized optical system (hardware layer) to preprocess complex spectral signals containing multiple alkali metals, ensuring high-quality, distortion-free multidimensional data is obtained on a single sensor. Subsequently, a pre-trained machine learning model (algorithm layer) is used to directly learn and model the complex nonlinear mapping relationship from this high-quality data to the actual concentration. This collaborative strategy of "optimized acquisition + intelligent analysis" fundamentally abandons the traditional path of relying on complex physical model calibration, thereby achieving high-precision and highly versatile real-time online detection of alkali metals released during the combustion of biomass and other fuels.

[0046] Example 1

[0047] This embodiment provides a visual detection system for alkali metals based on flame emission spectroscopy. (Refer to...) Figures 1 to 5 The system 1 includes a flame atomic emission system 100, a multispectral imaging system 200, and a processing unit 300.

[0048] A flame atomic emission system 100 is used for the thermal excitation of alkali metals in a sample. The system includes a burner 101 for generating a stable flame and a sample introduction device. In this embodiment, the burner 101 is an acetylene-air premixed flame burner. Acetylene is chosen as the fuel gas because its high-temperature flame provides a stable combustion environment and has low chemical interference, which helps improve the accuracy and sensitivity of the analytical results.

[0049] A multispectral imaging system 200 is used to capture the spectral radiation of alkali metals in a flame. (Reference) Figure 5 The system includes a single imaging sensor 201 and an optical component 202.

[0050] Selection and Design Basis of Characteristic Wavelengths: The first step in the design of this system is to determine the characteristic wavelengths of the alkali metals to be tested. Spectral analysis shows that sodium (Na), lithium (Li), potassium (K), and cesium (Cs) have unique and clear emission lines under flame excitation. To improve the signal-to-noise ratio, this invention selects the strongest emission lines of each element as the detection targets, specifically: the 588.31 nm line of Na, the 671.08 nm line of Li, the 766.58 nm line of K (whose intensity is significantly higher than that of the 770.19 nm line), and the 852.22 nm line of Cs (whose intensity is significantly higher than that of the 894.32 nm line).

[0051] Intensity Difference Problem and Solution: Under the same concentration and excitation conditions, the radiation intensity of different alkali metals varies greatly, with Na exhibiting the highest intensity, followed by Li and K, while Cs shows the weakest. This is determined by the inherent physical properties of each element, such as transition probabilities and ionization energies. To address this issue and avoid overexposure of strong signals (such as Na) while weak signals (such as Cs) remain undetectable in a single exposure, the optical component 202 innovatively incorporates a bandpass filter 2022 and an intensity filter 2023 in each optical channel 204. In a specific, preferred embodiment, the above-mentioned multispectral imaging system 200 is implemented as follows:

[0052] Camera and Structure: The imaging sensor 201 uses a 1 / 1.7-inch industrial camera (e.g., the MV-CE200-10UM model from Hijrobot, China), with a resolution of 4024×3036 pixels (sensor width 7.6mm, height 5.7mm) and a frame rate of 30 frames per second. The main structure 203 and the stage for fixing the optical components are 3D printed using PLA material.

[0053] Optical assembly: Optical assembly 202 includes four convex lenses 2021, each with a diameter of 3mm and a focal length of 3.2mm. It also includes four sets of filters, with each set containing a bandpass filter 2022 and an intensity filter 2023, both measuring 3.6mm × 3.6mm.

[0054] Filter configuration: Based on the aforementioned characteristic wavelengths and intensity differences, the specific filter configurations for the four channels are as follows:

[0055] First channel (Na): The center wavelength of bandpass filter 2022 is 590nm and the bandwidth is 40nm; in order to cope with the extremely strong spectral radiation of Na, the transmittance of intensity filter 2023 was experimentally determined to be 30%.

[0056] Second channel (Li): The center wavelength of the bandpass filter 2022 is 670nm and the bandwidth is 30nm; the transmittance of the intensity filter 2023 is determined to be 25%.

[0057] The third channel (K): the center wavelength of the bandpass filter 2022 is 760nm and the bandwidth is 35nm; the transmittance of the intensity filter 2023 is determined to be 40%.

[0058] Fourth channel (Cs): The center wavelength of the bandpass filter 2022 is 852nm and the bandwidth is 25nm; to ensure that the signal can be effectively detected, the transmittance of the intensity filter 2023 is determined to be 90%.

[0059] The processing unit 300 is configured to analyze image data acquired by the imaging sensor 201 and loads a pre-trained machine learning model, such as a two-dimensional convolutional neural network (2D-CNN), VGG-16, or ResNet-18. Its workflow includes: Image acquisition and segmentation: receiving data frames containing multiple spatially separated images from the imaging sensor 201 and segmenting them into regions of interest (ROIs) corresponding to each optical channel. Feature extraction: extracting at least one image feature for each ROI to form an input feature vector, which is a numerical array representing the light intensity information of each channel. Prediction output: inputting the feature vector to the pre-trained machine learning model (e.g., convolutional neural network CNN, support vector regression SVR, or gradient boosting tree XGBoost), the model directly outputs the predicted concentration of alkali metals.

[0060] Example 2

[0061] This embodiment further defines the system for liquid sample detection based on Embodiment 1. Sample preparation: The liquid sample to be tested can be a solution prepared from biomass raw materials. For example, 0.2g of biomass sample (such as wheat straw, rapeseed straw, corn cob, rice straw, or sawdust) can be weighed, 6mL of nitric acid can be added, and after pretreatment and microwave digestion, the volume can be adjusted to 100mL with deionized water to prepare the test solution.

[0062] Import device: Refer to Figure 2 The sample introduction device is a liquid atomization introduction system. This system includes an atomizer 102 (e.g., Beijing Haiguang GGX type), an atomization chamber 103, and a baffle 104. High-pressure air (e.g., using an optimized flow rate of 11 L / min) draws in the test solution and atomizes it into an aerosol. After premixing with acetylene gas in the atomization chamber 103, the aerosol is introduced into the burner 101.

[0063] Example 3

[0064] This embodiment further defines the system for detecting solid powder samples based on Embodiment 1.

[0065] Sample preparation: The solid powder sample to be tested can be a biomass raw material. For example, wheat straw, rapeseed straw, and other raw materials can be coarsely crushed, dried at 80℃ for 5 hours, further pulverized and sieved to obtain biomass powder with a particle size distribution of 100-200 mesh. In addition, different powders can be mixed in a 1:1 mass ratio to prepare a binary mixture sample.

[0066] Import device: Refer to Figure 3The sample introduction device is a powder pneumatic conveying system. This system includes a pneumatic conveying device consisting of a precision screw feeder 105 controlled by an STM32 microcontroller 106 and a vacuum generator 107. The high-precision stepper motor module of the screw feeder (e.g., screw pitch 1mm, positioning accuracy ±0.1mm) ensures that the powder is supplied at a stable rate, and the pneumatic conveying device uniformly feeds it into the burner 101.

[0067] Example 4

[0068] This embodiment further defines the system for detecting smoke samples based on Embodiment 1.

[0069] Sample source: Smoke may be generated by burning the biomass powder or mixture thereof prepared in Example 3 in a combustion chamber.

[0070] Import device: Refer to Figure 4 The system includes a smoke generation unit and an inlet unit. The smoke generation unit consists of a combustion chamber 108, a condensation device for cooling, and a smoke delivery vacuum pump 109. The inlet unit introduces the delivered smoke into the atomization chamber of the premixed burner, where it is mixed with acetylene and air for secondary excitation and analysis.

[0071] Example 5

[0072] This embodiment provides a method for optimizing detection conditions. Before sample testing, the method uses a schlieren imaging system 400 to visualize the flame flow field to assess the impact of different acetylene to air flow ratios on the test results.

[0073] Reference Figure 2 (b) The schlieren imaging system 400 serves as a powerful flow field diagnostic tool for visualizing flame structure, analyzing laminar and turbulent flow, and observing thermal plumes and density gradients. It consists of a coaxial point light source.

[0074] It consists of 401, a concave mirror, 402, a knife edge, 403, and a high-speed camera.

[0075] In one specific embodiment, the system operates and is configured as follows: Light emitted from a coaxial point light source 401, consisting of an LED and a beam splitter, is reflected by a concave mirror 402 (e.g., 203 mm in diameter, 750 mm in focal length) and focused onto a blade edge 403. The blade edge 403 blocks half of the focal point to enhance the contrast of the schlieren image. The flame region of the burner 101 is positioned in front of the concave mirror 402; changes in the density gradient within the flow field cause the incident light to deflect, thereby visualizing the flame's hydrodynamics. Finally, a high-speed camera 404 (e.g., an NPX-GS6500UM model, 1000 fps, 370 × 400 pixels resolution) captures the dynamic changes in the flow field.

[0076] The specific operation steps of this method are as follows: the operator adjusts the flow ratio of fuel gas (acetylene) and combustion gas (air) according to the visualized flow field image captured in real time by the high-speed camera 404 in order to obtain a stable laminar flame, which refers to a flame shape with clear flame boundaries, no obvious shaking or vortex, and smooth internal streamlines.

[0077] Example 6

[0078] This embodiment provides a method for detecting alkali metals based on multispectral imaging and machine learning, which includes the following core steps:

[0079] a) Signal acquisition: A multispectral imaging system 200 with multiple optical channels 204 is used to image the spectrum excited by the flame atomic emission system 100. Each optical channel 204 is pre-equalized and adjusted for the spectral radiation intensity of a specific alkali metal through an intensity filter 2023, and a data frame containing multiple spatially separated images is generated on a single imaging sensor 201.

[0080] b) Feature extraction: The processing unit 300 extracts at least one image feature from each spatially separated image of the data frame and combines them into an input feature vector.

[0081] c) Concentration prediction: The feature vector is input into a pre-trained machine learning model, which directly outputs the concentration of at least one alkali metal.

[0082] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A visual detection system for alkali metals based on flame emission spectroscopy, characterized in that, The system includes: A flame atomic emission system configured to thermally excite multiple alkali metals in a sample to emit spectral radiation, the flame atomic emission system further comprising a sample introduction device for introducing the sample into the excitation region; A multispectral imaging system configured to capture the spectral radiation, the multispectral imaging system comprising: A single imaging sensor; and An optical assembly includes multiple optical channels configured to simultaneously project the spectral radiation as multiple spatially separated images onto different regions of a single imaging sensor in a single exposure, and each optical channel includes: A bandpass filter is used to separate the characteristic wavelength of a specific alkali metal; and An intensity filter with a preset transmittance is used to equalize the spectral radiation intensity of different alkali metals; A processing unit configured for: Image data containing multiple spatially separated images is received from the single imaging sensor; and based on the preset correspondence between the multiple optical channels and the multiple spatially separated images, a pre-trained machine learning model is used to analyze each spatially separated image, wherein the machine learning model is used to establish a nonlinear mapping relationship between the image data and at least one feature of the multiple alkali metals, so as to achieve direct prediction of the feature.

2. The system according to claim 1, characterized in that, The sample is in liquid form, and the sample introduction device includes an atomizer for converting the liquid sample into an aerosol.

3. The system according to claim 1, characterized in that, The sample is in the form of a solid powder, and the sample introduction device includes a screw feeder and a pneumatic conveying device.

4. The system according to claim 1, characterized in that, The sample is in the form of smoke, and the sample introduction device includes a combustion chamber for generating the smoke and a vacuum pump for delivering the smoke.

5. The system according to claim 1, characterized in that, The plurality of alkali metals includes at least two selected from sodium, potassium, lithium and cesium.

6. The system according to claim 5, characterized in that, When the specific alkali metal is sodium, lithium, potassium and cesium, the transmittance of the intensity filters corresponding to the alkali metal is approximately 30%, 25%, 40% and 90%, respectively.

7. The system according to claim 1, characterized in that, The pre-trained machine learning model is a convolutional neural network.

8. A visual detection method for alkali metals based on flame emission spectroscopy, characterized in that, The method includes the following steps: Provide a flame for thermally activating alkali metals; The flow field dynamics characteristics of the flame are visualized and monitored using schlieren imaging technology; Based on the visualized monitoring results of the flow field dynamics, the combustion parameters of the flame are adjusted to obtain the flame flow field structure; and The sample to be tested is introduced into the flame flow field structure for flame emission spectroscopy detection.

Images