Element spectrum semi-quantitative analysis system

Abstract

The invention relates to the technical field of material analysis and detection, in particular to an element spectrum semi-quantitative analysis system which comprises an excitation device, a spectrum light splitting device and the excitation device are optically connected, the spectrum light splitting device comprises an optical grating, and an image acquisition device comprises an industrial camera and is optically connected to the spectrum light splitting device. The driving device is in transmission connection with the grating, the control unit is electrically connected with the driving device and the image acquisition device, the driving device controls the rotation angle of the grating through the control unit, the image acquisition device acquires spectral images at multiple angles and transmits the spectral images to the processing unit, and the processing unit is electrically connected with the processing unit. And the processing unit performs element type identification and content semi-quantitative analysis on the spectral image based on a deep learning model, so that classification identification and content prediction tasks of elements can be completed at the same time, and the accuracy of element identification and the reliability of semi-quantitative analysis are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of materials analysis and testing technology, specifically to a semi-quantitative elemental spectral analysis system. Background Technology

[0002] In many industrial fields such as metallurgy, machinery manufacturing, aerospace and quality supervision, the types and contents of metallic elements directly determine the mechanical properties, corrosion resistance and processing characteristics of materials. Therefore, the compositional analysis of metallic materials is a key link to ensure product quality and optimize production processes.

[0003] Current elemental spectral analysis mainly relies on spectroscopic techniques, where technicians observe the brightness of the spectral lines of the excited sample with their own eyes and perform semi-quantitative analysis by combining it with standard spectra. Therefore, this method has the following drawbacks: First, this operation depends on the operator's experience and judgment, and the human eye has limited ability to distinguish spectral lines, making it difficult to differentiate spectral lines of similar wavelengths, which leads to a significant increase in the probability of human error. In addition, each analysis requires manual adjustment of the grating angle and observation of each line individually, which is cumbersome and inefficient. Summary of the Invention

[0004] The purpose of this invention is to provide a semi-quantitative elemental spectral analysis system to improve detection accuracy while increasing detection efficiency to a limited extent.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a semi-quantitative elemental spectral analysis system, comprising: Excitation device; A spectroscopic spectrometer, optically connected to the excitation device, the spectroscopic spectrometer including a grating; The image acquisition device is optically connected to the spectral dispersive device; The driving device is connected to the grating drive; The control unit is electrically connected to both the drive device and the image acquisition device. The processing unit is signal-connected to the image acquisition device; The driving device controls the rotation angle of the grating through the control unit, the image acquisition device acquires spectral images from multiple angles and transmits them to the processing unit, and the processing unit performs element identification and semi-quantitative content analysis on the spectral images based on a deep learning model.

[0006] In some embodiments, the processing unit identifies the element types in the spectral image based on a deep learning model, specifically by outputting the element types present in the sample to be tested through the deep learning model.

[0007] In some embodiments, the processing unit performs semi-quantitative content analysis on the spectral image based on a deep learning model, specifically including outputting an estimated content value of at least one element in the sample to be tested through the deep learning model.

[0008] In some embodiments, the deep learning model is a convolutional neural network model.

[0009] In some embodiments, the convolutional neural network model includes multiple convolutional layers, pooling layers, and fully connected layers. The fully connected layers include a Softmax activation function output branch for element type identification and a linear regression output branch for semi-quantitative content analysis.

[0010] In some embodiments, a display unit is further included, which is signal-connected to the processing unit and is used to display element identification and semi-quantitative analysis results.

[0011] In some embodiments, the deep learning model training process employs a joint loss function, which includes cross-entropy loss for element classification and mean squared error loss for element concentration prediction.

[0012] In some embodiments, the excitation device includes a high-voltage power supply and an arc excitation electrode.

[0013] In some embodiments, the driving device includes a stepper motor, a coupling, and a gear. The coupling is fixedly connected to the grating, and the coupling is rotatably connected to the stepper motor through the gear. The motor shaft of the stepper motor is used to drive the gear to rotate.

[0014] In some embodiments, the grating has symmetrically arranged entrance slits and exit slits, and the industrial camera is disposed at the exit slit of the grating. The width of the entrance slit and the exit slit is 0.1-0.5 mm.

[0015] Furthermore, the beneficial effects of the present invention are as follows: This invention integrates an excitation device, a spectral dispersive device, an image acquisition device, a driving device, a control unit, and a processing unit. The processing unit also incorporates a deep learning model. This model integrates global features and local intensity information from the spectral image, enabling simultaneous element classification and content prediction. This effectively improves the accuracy of element identification and the reliability of semi-quantitative analysis, thereby reducing human error and increasing analytical precision. Meanwhile, this invention uses a stepper motor-driven grating, which automatically adjusts the grating by controlling the rotation angle of the stepper motor, thereby acquiring spectral images of different wavelength ranges. This achieves automation, precision, and continuity of spectral scanning, significantly improving analysis efficiency and spectral resolution, and is suitable for rapid identification of various elements. Attached Figure Description

[0016] Figure 1 The overall structural framework diagram of the elemental spectral semi-quantitative analysis system provided by the present invention; Figure 2 This is a schematic diagram showing the cooperation between the image acquisition device and the driving device in the elemental spectral semi-quantitative analysis system provided by the present invention.

[0017] In the diagram: 1-Industrial camera, 2-Stepper motor, 3-Gear, 4-Synchronous belt, 5-Coupling, 6-Raster. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In the description of the embodiments of the present invention, unless otherwise stated, " / " means "or", for example, A / B can mean A or B; "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships, for example, A and / or B can mean: A alone, A and B simultaneously, and B alone. In addition, in the description of the embodiments of the present invention, "multiple" means two or more. Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, features defined with "first" and "second" can explicitly or implicitly include one or more of that feature. In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more.

[0019] like Figures 1-2 As shown, an embodiment of the present invention provides a semi-quantitative elemental spectral analysis system, comprising: Excitation device; A spectroscopic spectrometer is optically connected to an excitation device, and the spectroscopic spectrometer includes a coupling 6; Image acquisition device, optically connected to spectral dispersive device; The drive unit is connected to the coupling 6 for transmission. The control unit is electrically connected to the drive unit and the image acquisition unit; The processing unit is connected to the image acquisition device via signals; The drive unit controls the rotation angle of the coupling 6 through the control unit, the image acquisition device acquires spectral images from multiple angles and transmits them to the processing unit, and the processing unit uses a deep learning model to identify the element types and perform semi-quantitative analysis of the content of the spectral images.

[0020] In the above structure, the excitation device consists of a high-voltage power supply and an electric arc excitation device, which is used to excite electric arc light on the metal sample. The high-voltage power supply is connected to the electrode device through wires, the electrode device is mechanically fixed to the sample stage, and the control unit controls the power switch through a relay to achieve controllable excitation of the electric arc.

[0021] In one possible implementation, the processing unit identifies element types in the spectral image based on a deep learning model. Specifically, this includes outputting the types of elements present in the sample through the deep learning model. The processing unit also performs semi-quantitative content analysis on the spectral image based on the deep learning model. Specifically, this includes outputting an estimated content value of at least one element in the sample through the deep learning model. The deep learning model is a convolutional neural network model, which includes multiple convolutional layers, pooling layers, and fully connected layers. The fully connected layers include a Softmax activation function output branch for element type identification and a linear regression output branch for semi-quantitative content analysis. The deep learning model training process uses a joint loss function, which includes cross-entropy loss for element classification and mean squared error loss for element concentration prediction.

[0022] After the spectral image acquisition is completed, the system first performs a unified image preprocessing operation to ensure image quality and standardize the input format. At the same time, it will shield the interference caused by shooting angle, brightness fluctuation or background light in the image. The specific process includes: (1) the image is extracted by default to the center 200×1500 pixel area to ensure that only the area containing effective spectral lines is retained; (2) the background interference is removed by the rolling ball algorithm with a sphere radius set to 50 pixels; (3) pixel value normalization is performed to unify the dynamic range of the model input; (4) the median filtering or average filtering method of five-frame sliding window is used to perform multi-frame image fusion processing to enhance the signal-to-noise ratio and improve the weak spectral line recognition ability; (5) all images are uniformly scaled to 256×64 pixels before being input into the model; (6) the data augmentation mechanism is enabled during the training phase, and the image preprocessing process is completed at this time.

[0023] After image preprocessing, the data is input into a deep learning recognition model for further processing. However, before the data input, the deep learning model in this invention employs a lightweight two-dimensional convolutional neural network structure. This network structure includes three convolutional modules: the first layer has a 3×3 kernel size, 32 channels, a stride of 1, and uses "same" padding, followed by batch normalization and ReLU activation, and then downsampling using 2×2 max pooling. The second layer is similar to the first, but the number of channels is increased to 64. The third convolutional layer expands to 128 channels, continuing with 3×3 convolution and batch normalization. After pooling and simplification operations, the convolution output is flattened into a 32768-dimensional vector and fed into a fully connected layer. The fully connected layer contains a hidden layer with ReLU activation function and Dropout set to 0.3 to prevent overfitting. The final output is divided into two task branches: one is to output the probability distribution of N element types using the Softmax activation function, where N is the number of element types supported by the model; the other is a linear regression branch, used to output the concentration prediction value of one or more target elements, with units that can be set to wt.%, ppm or other quantitative units, supporting linear regression prediction in the range of 0.01–10.

[0024] In addition, the model training uses the Adam optimizer with an initial learning rate of 0.001, which automatically decreases by 5% every 10 epochs. The loss function adopts a joint loss form: cross-entropy loss is used for element classification, and mean squared error is used for element concentration prediction. The final loss is a weighted sum of the two. The training data uses a batch size of 32 and the number of training epochs is set to 100. In each epoch, the Top-1 classification accuracy and regression RMSE are evaluated on the validation set. The typical training time is about 15-30 minutes in a GPU environment. For easy deployment, the trained model can be converted to ONNX or Tensor RT format for local inference on edge computing devices such as Jetson Nano, RK3588, Raspberry Pi + NPU extension module, etc.

[0025] During the deployment phase, the system supports real-time image input and local inference, with a single image analysis time of less than 0.1 seconds. The output includes the predicted element types and their concentration estimates. Simultaneously, during the detection process, the system provides a spectral self-calibration mechanism, which automatically performs positional offset correction and intensity normalization through preset standard sample spectral lines, ensuring good repeatability of the analysis results. In addition, the system can support simultaneous identification of multiple elements and output of multiple target concentrations, and has good scalability and application adaptability, making it suitable for scenarios such as metallurgical composition analysis, material sorting, emergency detection, teaching and research.

[0026] One possible implementation also includes a display unit that is signal-connected to the processing unit for displaying element identification and semi-quantitative analysis results.

[0027] In one possible implementation, the driving device includes a stepper motor 2, a coupling 5, and a gear 3. The coupling 5 is fixedly connected to a coupling 6, and the coupling 5 is rotatably connected to the stepper motor 2 through the gear 3. The motor shaft of the stepper motor 2 is used to drive the gear 3 to rotate. The coupling 6 has symmetrically arranged entrance slits and exit slits. The industrial camera 1 is set at the exit slit of the coupling 6. The width of the entrance slit and the exit slit is 0.1-0.5 mm. In the above structure, this application uses an entrance slit with a width of 0.1-0.5 mm to limit the width of the incident light, while the exit slit is symmetrical to the entrance slit to limit the emitted light. Furthermore, in this embodiment, the industrial camera 1 is a color CMOS area array industrial camera 1 with a resolution ≥1280×1024 and a frame rate ≥30fps. It is equipped with an 8mm focal length lens, and its field of view matches the exit slit of the coupling 6 to ensure that the spectral lines enter the field of view completely. The industrial camera 1 is connected to the image acquisition device via a USB interface. During detection, the coupling 6 and the stepper motor 2 are connected via a synchronous belt 4 to achieve precise angle control. However, in the above process, through the action of the control unit, the step angle of the stepper motor 2 is 1.8° / step, with an accuracy of 0.1°, corresponding to a spectral drift of less than 0.1nm, ensuring the fineness and continuity of spectral acquisition, and ensuring that high-quality spectral images can be obtained at each angular position of the coupling 6. The core control of the control unit adopts an STM32 or Raspberry Pi embedded platform, which is responsible for driving the stepper motor 2, synchronously triggering the camera to capture images, and completing data caching and image stitching processing to achieve automatic identification and semi-quantitative analysis of the types and contents of elements in the target sample. Its hardware uses NVIDIA Jetson Nano as the main controller.

[0028] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A semi-quantitative elemental spectral analysis system, characterized in that, include: Excitation device; A spectroscopic spectrometer, optically connected to the excitation device, the spectroscopic spectrometer including a grating; An image acquisition device, including an industrial camera, is optically connected to the spectral dispersive device; The driving device is connected to the grating drive; The control unit is electrically connected to both the drive device and the image acquisition device. The processing unit is signal-connected to the image acquisition device; The driving device controls the rotation angle of the grating through the control unit, the image acquisition device acquires spectral images from multiple angles and transmits them to the processing unit, and the processing unit performs element identification and semi-quantitative content analysis on the spectral images based on a deep learning model.

2. The elemental spectral semi-quantitative analysis system as described in claim 1, characterized in that, The processing unit identifies the element types in the spectral image based on a deep learning model, specifically by outputting the element types present in the sample to be tested through the deep learning model.

3. The elemental spectral semi-quantitative analysis system as described in claim 2, characterized in that, The processing unit performs semi-quantitative content analysis on the spectral image based on a deep learning model, specifically including outputting an estimated content value of at least one element in the sample to be tested through the deep learning model.

4. The elemental spectral semi-quantitative analysis system as described in claim 3, characterized in that, The deep learning model is a convolutional neural network model.

5. The elemental spectral semi-quantitative analysis system as described in claim 4, characterized in that, The convolutional neural network model includes multiple convolutional layers, pooling layers, and fully connected layers. The fully connected layers include a softmax activation function output branch for element identification and a linear regression output branch for semi-quantitative content analysis.

6. The elemental spectral semi-quantitative analysis system as described in claim 1, characterized in that, It also includes a display unit, which is signal-connected to the processing unit and is used to display element identification and semi-quantitative analysis results.

7. The elemental spectral semi-quantitative analysis system as described in claim 2, characterized in that, The deep learning model training process adopts a joint loss function, which includes cross-entropy loss for element classification and mean squared error loss for element concentration prediction.

8. The elemental spectral semi-quantitative analysis system as described in claim 1, characterized in that, The excitation device includes a high-voltage power supply and an arc excitation electrode.

9. The elemental spectral semi-quantitative analysis system as described in claim 1, characterized in that, The driving device includes a stepper motor, a coupling, and a gear. The coupling is fixedly connected to the grating, and the coupling is rotatably connected to the stepper motor through the gear. The motor shaft of the stepper motor is used to drive the gear to rotate.

10. The elemental spectral semi-quantitative analysis system as described in claim 9, characterized in that, The grating has symmetrically arranged entrance slits and exit slits, and the industrial camera is set at the exit slit of the grating. The width of the entrance slit and the exit slit is 0.1-0.5 mm.

Images