A multimodal sensing adaptive control system and method for pyrolysis parameters of printed circuit boards

Abstract

This invention belongs to the field of adaptive control technology for pyrolysis parameters. It discloses a multimodal sensing adaptive control system and method for pyrolysis parameters of printed circuit boards (PCBs). The method includes: extracting features from the spectral signals of R-block PCBs to obtain a set of fluorescence spectral feature vectors; extracting features from the reflectance spectral data of R-block PCBs to obtain a set of near-infrared reflectance spectral feature vectors; extracting features from the surface image of R-block PCBs to obtain a set of PCB image feature vectors; evaluating R-block PCBs based on the fluorescence spectral feature vector set, the near-infrared reflectance spectral feature vector set, and the PCB image feature vector set to obtain the predicted manufacturing age and aging assessment level of R-block PCBs; and intelligently adaptively controlling the pyrolysis parameters of the PCBs. This application enables the pyrolysis parameters of PCBs to accurately match the decomposition requirements of PCBs under different manufacturing backgrounds and aging conditions, improving the precision of pyrolysis parameter control.

Description

Technical Field

[0001] This invention relates to the field of adaptive control technology for pyrolysis parameters of printed circuit boards, and more specifically, to a multimodal sensing adaptive control system and method for pyrolysis parameters of printed circuit boards. Background Technology

[0002] With the rapid development of the electronics and information industry, the number of discarded circuit boards is increasing daily. Pyrolysis, as a recycling pathway, can effectively achieve a circular economy and environmental protection because it can efficiently convert organic matter into combustible gas and simultaneously enrich metal components. To improve the energy utilization rate and processing quality of the pyrolysis process, existing technologies typically control the process based on preset temperature curves, heating rates, residence times, and atmosphere ratios. Some schemes adjust the furnace temperature and gas flow rate through infrared thermography or tail gas concentration feedback to achieve preliminary closed-loop process control.

[0003] However, in actual processing, the raw materials for circuit boards are diverse, and the metals, resins, fillers, and even welding processes used in different manufacturing eras vary significantly. Furthermore, aging due to prolonged use further exacerbates compositional fluctuations, making it difficult for traditional uniform process parameters to accommodate all raw material types. This leads to frequent problems such as inaccurate parameter adaptation, incomplete reactions, and high energy consumption. In addition, most existing technologies rely on operators setting parameters based on experience, lacking intelligent identification of the actual properties of raw materials and the ability to adapt parameters accordingly, making it difficult to achieve precise and real-time control of circuit board parameters. Especially during batch mixing or raw material switching, inaccurate settings often lead to process risks such as coking, unstable gas composition, or excessive dioxin emissions, hindering the widespread application of pyrolysis treatment equipment.

[0004] Therefore, there is an urgent need for a new method for controlling the pyrolysis of circuit boards that integrates raw material property perception and intelligent parameter setting, in order to address the challenges of resource utilization of highly heterogeneous circuit board raw materials. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a multimodal sensing adaptive control method for circuit board pyrolysis parameters, comprising:

[0006] The spectral signals of the R-block circuit board acquired in the fluorescence spectral acquisition area are analyzed and their features are extracted to obtain a set of fluorescence spectral feature vectors;

[0007] Feature extraction and structuring are performed on the reflectance spectral data of the R-block circuit board collected in the near-infrared reflectance spectral acquisition area to obtain a set of near-infrared reflectance spectral feature vectors;

[0008] Feature extraction processing is performed on the surface images of R circuit boards acquired by the image acquisition device to obtain a set of circuit board image feature vectors;

[0009] The R-block circuit boards are evaluated based on the fluorescence spectral feature vector set, the near-infrared reflectance spectral feature vector set, and the circuit board image feature vector set to obtain the predicted manufacturing age and aging assessment level of the R-block circuit boards.

[0010] Based on the set of fluorescence spectral feature vectors, the set of near-infrared reflectance spectral feature vectors, the set of circuit board image feature vectors, the predicted manufacturing age and aging assessment level of R circuit boards, pyrolysis parameter adjustment instructions are generated, and intelligent adaptive control of the circuit board pyrolysis parameters is performed.

[0011] Furthermore, methods for intelligent adaptive control of circuit board pyrolysis parameters include:

[0012] Based on the predicted manufacturing age and aging assessment level of R-block circuit boards, a set of binary pairs of predicted manufacturing age and a set of binary pairs of aging assessment level are constructed.

[0013] The dominant component was determined and the dominant label value was constructed based on the binary set of predicted manufacturing age and the binary set of aging assessment level.

[0014] The dominant label value, fluorescence spectral feature vector set, near-infrared reflectance spectral feature vector set, and circuit board image feature vector set are input into the pyrolysis parameter setting model to obtain the corresponding pyrolysis parameter combination. The pyrolysis parameter combination includes the target pyrolysis temperature, heating rate, pyrolysis residence time, oxygen to inert gas ratio, and initial pressure of the pyrolysis reaction chamber.

[0015] The system control platform sends the pyrolysis parameter combination to the corresponding control equipment and executes it.

[0016] Furthermore, the method for obtaining the set of binary pairs for predicted manufacturing age and the set of binary pairs for aging assessment level includes:

[0017] S400: Let r be initially 1, and the range of r is from 1 to R;

[0018] S401: Obtain the predicted manufacturing age and aging assessment level of the r-th circuit board;

[0019] Determine whether the predicted manufacturing year already exists in the predicted manufacturing year tuple set; if it exists, increment the circuit board count value of the predicted manufacturing year tuple by 1; if it does not exist, construct a predicted manufacturing year tuple from the predicted manufacturing year of the r-th circuit board and 1, and add it to the predicted manufacturing year tuple set; an example of the predicted manufacturing year tuple is (predicted manufacturing year of the r-th circuit board, circuit board count value).

[0020] Determine whether the aging assessment level already exists in the aging assessment level binary set; if it exists, increment the circuit board count value of the aging assessment level binary set by 1; if it does not exist, construct an aging assessment level binary set by the aging assessment level of the r-th circuit board and 1, and add it to the aging assessment level binary set; an example of the aging assessment level binary set is (aging assessment level of the r-th circuit board, circuit board count value).

[0021] S402: Let r = r + 1. If r is less than or equal to R, return to S401 and continue execution; if r is greater than R, end the current process.

[0022] Furthermore, methods for determining the dominant component and constructing the dominant label value based on the predicted manufacturing year binary set and the aging assessment level binary set include:

[0023] Initialize the dominant identifier for predicting the manufacturing year and the dominant identifier for the aging assessment level to No;

[0024] The circuit board quantity count value of each predicted manufacturing year binary set is divided with R to obtain the corresponding circuit board quantity percentage, and the first quantity percentage set is constructed.

[0025] The circuit board quantity count value of each aging assessment level binary set in the aging assessment level binary set is divided with R to obtain the corresponding circuit board quantity percentage, and a second quantity percentage set is constructed.

[0026] Select the largest first quantity percentage from the first quantity percentage set, and determine whether the first quantity percentage is greater than the preset first quantity percentage threshold. If the determination result is yes, then set the dominant identifier of the predicted manufacturing year to yes.

[0027] Select the largest second quantity percentage from the second quantity percentage set, and determine whether the second quantity percentage is greater than the preset second quantity percentage threshold. If the determination result is yes, then set the dominant indicator of the aging assessment level to yes.

[0028] The dominant identifiers for predicting the manufacturing age and aging assessment level are matched with a pre-constructed dominant label value mapping table to obtain the dominant label values.

[0029] Furthermore, the methods for obtaining the predicted manufacturing age and aging assessment level of the R-block circuit board include:

[0030] S300: Let r be initially 1, and the range of r is from 1 to R;

[0031] S301: Obtain the fluorescence spectrum feature vector of the r-th circuit board from the fluorescence spectrum feature vector set, obtain the near-infrared reflectance spectrum feature vector of the r-th circuit board from the near-infrared reflectance spectrum feature vector set, and obtain the circuit board image feature vector of the r-th circuit board from the circuit board image feature vector set.

[0032] S302: Input the fluorescence spectral feature vector, near-infrared reflectance spectral feature vector and circuit board image feature vector of the r-th circuit board into the comprehensive evaluation model to obtain the predicted manufacturing age and aging assessment level of the r-th circuit board.

[0033] S303: Let r = r + 1. If r is less than or equal to R, return to S301 to continue execution. If r is greater than R, end the current process.

[0034] Furthermore, the method for obtaining the set of fluorescence spectral feature vectors includes:

[0035] The X-ray fluorescence spectroscopy acquisition device emits a micro-focus X-ray beam with a focal spot size of less than 100 micrometers onto the surface of the R-block circuit board; under the bombardment of X-rays, the inner-shell electrons of the elements on the surface of the circuit board undergo transitions and instantaneously release X-ray fluorescence with characteristic energy;

[0036] After X-ray fluorescence enters the spectrometer detector, each fluorescent photon of the X-ray fluorescence forms an independent incident photon event. The spectrometer detector collects and converts the electronic signals released by the incident photon events, generating a pulsed voltage signal whose amplitude is linearly related to the fluorescence energy.

[0037] After being pre-amplified, shaped and converted from analog to digital, the pulsed voltage signal enters the multichannel analyzer for energy channel classification and statistical accumulation. The spectral analysis software automatically extracts the peak energy, peak height and relative ratio of each element in the R-block circuit board.

[0038] The peak energy, peak intensity, and relative ratio of the R circuit boards are normalized and vectorized to construct R sets of fluorescence spectral feature vectors, which are then used to construct a set of fluorescence spectral feature vectors.

[0039] Furthermore, the method for obtaining the near-infrared reflectance spectral feature vector set includes:

[0040] The near-infrared reflectance spectrometer emits near-infrared light with a wavelength range of 900 to 1700 nanometers onto the surface of R circuit boards. The near-infrared detector receives the near-infrared light signal reflected back from the surface of the circuit boards and samples it according to the number of wavelength points Q. That is, the reflectance of each circuit board is collected at Q wavelength points to form a Q-dimensional reflectance vector.

[0041] Principal component analysis is performed on the Q-dimensional reflectance vector corresponding to each circuit board to obtain the near-infrared reflectance spectral feature vectors corresponding to R circuit boards; the near-infrared reflectance spectral feature vectors corresponding to R circuit boards are then used to construct a set of near-infrared reflectance spectral feature vectors.

[0042] Furthermore, the method for obtaining the near-infrared reflectance spectral feature vector corresponding to the R-block circuit board includes:

[0043] Construct a reflectivity matrix from the Q-dimensional reflectivity vector corresponding to each circuit board;

[0044] Subtract the mean reflectance of each column from the reflectance data in the reflectance matrix to obtain a zero-mean matrix. Calculate the covariance matrix based on the zero-mean matrix. The covariance matrix is ​​used to measure the linear correlation between reflectance values ​​of different bands.

[0045] Find the eigenvalues ​​of the covariance matrix to obtain W eigenvalues. Find the corresponding eigenvector for each eigenvalue. Sort the W eigenvalues ​​in descending order to form an eigenvalue set. Then, construct an eigenvector set from the eigenvalue set and the corresponding eigenvectors of the eigenvalue set.

[0046] The number of principal components (ZCF) is determined based on the eigenvalue set and a preset cumulative contribution rate threshold.

[0047] The first ZCF eigenvectors are selected from the eigenvector set to construct the principal component vector set;

[0048] Perform linear projection operations on the Q-dimensional reflectance vector of the R-block circuit board with the principal component vector set to obtain the principal component projection value sequence corresponding to the R-block circuit board. The principal component projection value sequence is the near-infrared reflectance spectral feature vector of the circuit board.

[0049] Furthermore, the method for determining the number of principal components (ZCF) based on the eigenvalue set and a preset cumulative contribution rate threshold includes:

[0050] Step 1: Sum the eigenvalues ​​in the eigenvalue set to obtain the total eigenvalues; set the initial value of the principal component number ZCF to 1;

[0051] Step 2: Select the first ZCF eigenvalues ​​from the eigenvalue set and sum them to obtain the cumulative eigenvalue. Divide the cumulative eigenvalue by the sum of the eigenvalues ​​to obtain the cumulative contribution rate.

[0052] Step 3: If the cumulative contribution rate is greater than the cumulative contribution rate threshold, the number of principal components ZCF is obtained, and the current process ends; if the cumulative contribution rate is less than or equal to the cumulative contribution rate threshold, ZCF = ZCF + 1 is set, and the process returns to step 2 to continue execution.

[0053] Furthermore, the method for obtaining the principal component projection value sequence corresponding to the R-block circuit board includes:

[0054] S100: Construct a principal component matrix from the set of principal component vectors; the principal component matrix has Q rows and ZCF columns; each column in the principal component matrix corresponds to a principal component vector; let the initial value of r be 1, and the value of r range from 1 to R;

[0055] S101: Obtain the element of the r-th row of the zero-mean matrix, denoted as the centered reflectivity vector;

[0056] S102: Multiply the centered reflectivity vector by the principal component matrix to obtain the sequence of principal component projection values;

[0057] S103: Let r = r + 1. If r is less than or equal to R, return to S101 and continue execution. If r is greater than R, end the current process.

[0058] Furthermore, the method for obtaining the circuit board image feature vector set includes:

[0059] S200: Let r be initially 1, and the range of r is from 1 to R;

[0060] S201: Preprocess the surface image of the r-th circuit board to obtain a preprocessed surface image of the r-th circuit board; the preprocessing includes brightness normalization, contrast enhancement, edge smoothing and artifact elimination;

[0061] S202: Divide the preprocessed surface image of the r-th circuit board into H×H equal-area image grids, extract the number of circuit board solder joints in each image grid, and calculate the solder joint distribution uniformity factor of the r-th circuit board; mark the image grids with a number of circuit board solder joints greater than a preset threshold as welding areas, and count the number of welding areas as the number of welding areas; extract the area of ​​the solder joint area and the area of ​​the image grid corresponding to each image grid, and calculate the average solder joint density of the r-th circuit board;

[0062] S203: Construct the circuit board image feature vector of the r-th circuit board from the solder joint distribution uniformity factor, the number of solder areas and the average solder joint density, and add it to the circuit board image feature vector set.

[0063] S204: Let r = r + 1. If r is less than or equal to R, return to S201 and continue execution. If r is greater than R, end the current process.

[0064] A multimodal sensing adaptive control system for PCB pyrolysis parameters, used to implement the aforementioned multimodal sensing adaptive control method for PCB pyrolysis parameters, includes:

[0065] The first processing module is used to analyze and extract features from the spectral signals of the R-block circuit board acquired in the fluorescence spectral acquisition area to obtain a set of fluorescence spectral feature vectors.

[0066] The second processing module is used to extract features and perform structured processing on the reflectance spectral data of the R-block circuit board collected in the near-infrared reflectance spectral acquisition area to obtain a set of near-infrared reflectance spectral feature vectors.

[0067] The third processing module is used to perform feature extraction processing on the surface image of the R-block circuit board acquired by the image acquisition device to obtain a set of circuit board image feature vectors.

[0068] The comprehensive evaluation module evaluates the R circuit boards based on the fluorescence spectral feature vector set, the near-infrared reflectance spectral feature vector set, and the circuit board image feature vector set, and obtains the predicted manufacturing age and aging assessment level of the R circuit boards.

[0069] The parameter control module generates pyrolysis parameter adjustment instructions based on the set of fluorescence spectral feature vectors, the set of near-infrared reflectance spectral feature vectors, the set of circuit board image feature vectors, the predicted manufacturing age and aging assessment level of R circuit boards, and performs intelligent adaptive control of the circuit board pyrolysis parameters.

[0070] Compared with existing technologies, the technical effects and advantages of the multimodal sensing adaptive control system and method for pyrolysis parameters of circuit boards of the present invention are as follows:

[0071] This invention provides a multimodal sensing adaptive control system and method for pyrolysis parameters of printed circuit boards. By fusing multi-source heterogeneous sensing information such as fluorescence spectral features, near-infrared reflectance spectral features and image features, and combining the predicted manufacturing age and aging assessment level of the printed circuit board, a dominant label value is constructed. Based on the dominant label value, a pyrolysis parameter setting model is driven to intelligently generate a combination of pyrolysis parameters, including the target pyrolysis temperature, heating rate, pyrolysis residence time, oxygen and inert gas ratio, and initial pressure of the pyrolysis reaction chamber.

[0072] Furthermore, this application enables dynamic adjustment of the multi-physical quantity control components of the pyrolysis device, allowing the pyrolysis process of circuit boards to precisely match the decomposition requirements of circuit boards under different manufacturing backgrounds and aging conditions, significantly improving the reaction efficiency and control stability of the pyrolysis process. Compared with the existing technology that uses uniform process parameters or relies on manual experience to set them, this invention realizes a data-driven intelligent adaptive parameter setting mechanism, effectively solving problems such as inaccurate parameter adaptation, excessive energy consumption, and incomplete reaction in the processing of diverse circuit boards. It effectively improves the accuracy and real-time performance of circuit board parameter control, demonstrating strong technological advancement and industrial practical value. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of a multimodal sensing adaptive control system for circuit board pyrolysis parameters according to Embodiment 1 of the present invention;

[0074] Figure 2 This is a flowchart of a multimodal sensing adaptive control method for pyrolysis parameters of a circuit board according to Embodiment 2 of the present invention;

[0075] Figure 3 Here is a flowchart of the feature extraction method for R-block circuit boards;

[0076] Figure 4 Flowchart of the method for obtaining the main tag value;

[0077] Figure 5 This is a schematic diagram of the data-driven input and control linkage for the pyrolysis parameters of the circuit board. Detailed Implementation

[0078] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0079] Example 1:

[0080] Please see Figure 1 As shown in the figure, this embodiment discloses a multimodal sensing circuit board pyrolysis parameter adaptive control system, including a first processing module, a second processing module, a third processing module, a comprehensive evaluation module, and a parameter control module. Each module is connected by wires and / or wirelessly to realize data transmission.

[0081] The first processing module is used to analyze and extract features from the spectral signals of R circuit boards acquired in the fluorescence spectral acquisition area to obtain a set of fluorescence spectral feature vectors. R is the total number of circuit boards processed in the current batch.

[0082] The method for obtaining the set of fluorescence spectral feature vectors includes:

[0083] The X-ray fluorescence spectroscopy acquisition device emits a micro-focus X-ray beam with a focal spot size of less than 100 micrometers onto the surface of the R-block circuit board to ensure accurate positioning of the excitation region and sufficient energy density. Under X-ray bombardment, the inner-shell electrons of the elements on the surface of the circuit board undergo transitions and instantaneously release X-ray fluorescence with characteristic energy.

[0084] After X-ray fluorescence enters the spectrometer detector, which is preferably a silicon drift detector (SDD) or a PIN diode detector, each fluorescent photon of the X-ray fluorescence forms an independent incident photon event. The spectrometer detector collects and converts the electronic signal released by the incident photon event to generate a pulsed voltage signal whose amplitude is linearly related to the fluorescence energy.

[0085] After being pre-amplified, shaped and converted from analog to digital, the pulsed voltage signal enters the multichannel analyzer for energy channel classification and statistical accumulation. The spectral analysis software automatically extracts the peak energy (peak center position), peak height (peak intensity) and relative ratio of each element in the R-block circuit board.

[0086] The peak energies, peak intensities, and relative ratios of R circuit boards are normalized and vectorized to construct R sets of fluorescence spectral feature vectors. These R sets of fluorescence spectral feature vectors are then used to construct a set of fluorescence spectral feature vectors. This set of fluorescence spectral feature vectors is used to comprehensively characterize the metal composition and abundance of the circuit boards, providing highly reliable spectroscopic input for determining the manufacturing age, identifying alloy types, and estimating metal load.

[0087] It should be noted that the X-ray beam refers to a spatially concentrated radiation energy flow continuously generated from an X-ray emission tube and oriented by a collimator, possessing a definite radiation direction and energy density distribution. The microfocus refers to a focal point with a diameter no greater than 100 micrometers formed when the electron beam inside the X-ray source bombards the anode target, thus creating an X-ray irradiation region with an extremely small focal spot size on the sample surface, enabling micro-area excitation and high-resolution spectroscopic analysis. By irradiating the target circuit board surface with a microfocus X-ray beam, the characteristic fluorescence of specific material elements can be excited within a limited range, improving the accuracy of spectral acquisition and spatial resolution.

[0088] By analyzing the X-ray fluorescence signal generated on the surface of the circuit board and extracting its features, the obtained fluorescence spectral feature vector can characterize the types, contents, and relative proportions of metal elements contained in the circuit board. This has the following significant technical benefits and comprehensive application value: First, the peak energy in the fluorescence spectral feature vector can accurately identify various metal elements in the circuit board surface, such as Sn (tin), Pb (lead), Cu (copper), and Br (bromine), thus achieving consistency in determining the manufacturing age and process system (e.g., whether it is lead-free solder). Second, the peak intensity and relative ratio can reflect the distribution abundance of each metal element in the circuit board, which can be used to estimate the total alloy load level of the entire board, providing a basis for subsequent pyrolysis parameter control. Third, as a highly structured and quantifiable material characterization vector, the fluorescence spectral feature vector has advantages such as non-contact, speed, and high resolution. It can achieve online classification and difference identification of mixed batches of circuit board raw materials without interrupting the production cycle, thereby improving the intelligence level, adaptability, and resource utilization of the entire processing line. Compared to traditional raw material classification methods that rely on visual or manual identification, the fluorescence spectral feature vector introduced in this invention has stronger resistance to pollution interference and higher classification accuracy, making it particularly suitable for efficient and intelligent identification of electronic waste circuit boards with complex structures and significant differences in composition.

[0089] It should be noted that the peak energy, peak intensity, and relative ratio of each element are shown in Table 1, Element Data Table:

[0090] Table 1 Element Data Table

[0091]

[0092] The second processing module is used to extract features and structure the reflectance spectral data of the R-block circuit board collected in the near-infrared reflectance spectral acquisition area, resulting in a set of near-infrared reflectance spectral feature vectors. These feature vectors are used to characterize the circuit board's material composition, aging state, and organic residues.

[0093] The method for obtaining the near-infrared reflectance spectral feature vector set includes:

[0094] A near-infrared reflectance spectrometer emits near-infrared light with a wavelength range of 900 to 1700 nanometers onto the surface of R circuit boards. A near-infrared detector (preferably an InGaAs linear array detector) receives the near-infrared light signal reflected back from the surface of the circuit boards and samples it according to the number of wavelength points Q (e.g., if sampling is performed every 5 nanometers between 900 and 1700 nanometers, then the number of wavelength points is 161). That is, the reflectance of each circuit board is collected at Q wavelength points, forming a Q-dimensional reflectance vector.

[0095] Principal component analysis is performed on the Q-dimensional reflectance vector corresponding to each circuit board to obtain the near-infrared reflectance spectral feature vectors corresponding to R circuit boards; the near-infrared reflectance spectral feature vectors corresponding to R circuit boards are then used to construct a set of near-infrared reflectance spectral feature vectors.

[0096] The methods for obtaining the near-infrared reflectance spectral feature vector corresponding to the R-block circuit board include:

[0097] A reflectivity matrix is ​​constructed from the Q-dimensional reflectivity vector corresponding to each circuit board. The reflectivity matrix has R rows and Q columns. Each row of the reflectivity matrix corresponds to the reflectivity value of a circuit board at each wavelength point. The mean reflectivity of each column is subtracted from the reflectivity data in each column of the reflectivity matrix to obtain a zero-mean matrix. The covariance matrix is ​​calculated based on the zero-mean matrix. The covariance matrix is ​​used to measure the linear correlation between reflectivity in different bands.

[0098] Find the eigenvalues ​​of the covariance matrix to obtain W eigenvalues. Find the corresponding eigenvector for each eigenvalue. Sort the W eigenvalues ​​in descending order to form an eigenvalue set. Then, construct an eigenvector set from the eigenvalue set and the corresponding eigenvectors of the eigenvalue set.

[0099] The number of principal components (ZCF) is determined based on the eigenvalue set and a preset cumulative contribution rate threshold.

[0100] The first ZCF eigenvectors are selected from the eigenvector set to construct the principal component vector set;

[0101] Perform linear projection operations on the Q-dimensional reflectance vector of the R-block circuit board with the principal component vector set to obtain the principal component projection value sequence corresponding to the R-block circuit board. The principal component projection value sequence is the near-infrared reflectance spectral feature vector of the circuit board.

[0102] Methods for determining the number of principal components (ZCF) based on the eigenvalue set and a preset cumulative contribution rate threshold include:

[0103] Step 1: Sum the eigenvalues ​​in the eigenvalue set to obtain the total eigenvalues; set the initial value of the principal component number ZCF to 1;

[0104] Step 2: Select the first ZCF eigenvalues ​​from the eigenvalue set and sum them to obtain the cumulative eigenvalue. Divide the cumulative eigenvalue by the sum of the eigenvalues ​​to obtain the cumulative contribution rate.

[0105] Step 3: If the cumulative contribution rate is greater than the cumulative contribution rate threshold, the number of principal components ZCF is obtained, and the current process ends; if the cumulative contribution rate is less than or equal to the cumulative contribution rate threshold, ZCF = ZCF + 1, and the process returns to Step 2 to continue execution. For example, in this application, the cumulative contribution rate threshold can be set to 0.95.

[0106] The methods for obtaining the principal component projection value sequence corresponding to the R-block circuit board include:

[0107] S100: Construct a principal component matrix from the set of principal component vectors; the principal component matrix has Q rows and ZCF columns; each column in the principal component matrix corresponds to a principal component vector; let the initial value of r be 1, and the value of r range from 1 to R;

[0108] S101: Obtain the element of the r-th row of the zero-mean matrix, denoted as the centered reflectivity vector;

[0109] S102: Multiply the centered reflectivity vector by the principal component matrix to obtain the sequence of principal component projection values;

[0110] The calculation of the principal component projection value sequence is as follows:

[0111] ;

[0112] in, Let be the sequence of principal component projection values ​​corresponding to the r-th circuit board. The element in the r-th row of the zero-mean matrix is ​​the centered reflectivity vector corresponding to the r-th circuit board. Principal component matrix; The element in the first column of the principal component matrix. The element in the ZCF column of the principal component matrix.

[0113] S103: Let r = r + 1. If r is less than or equal to R, return to S101 and continue execution. If r is greater than R, end the current process.

[0114] It should be noted that the surface of a circuit board typically consists of an epoxy resin matrix, a glass fiber layer, a tin-based alloy soldering area, and residual flux or thermally conductive adhesive from the process. Each component exhibits specific absorption characteristics for different wavelengths of light in the near-infrared band. By standardizing, performing principal component analysis and principal component projection processing on near-infrared reflectance data from multiple circuit boards in the 900-1700 nm band, the resulting set of near-infrared reflectance spectral feature vectors constitutes an important component of the multimodal raw material fingerprinting system in this application. It plays a crucial role in multiple stages, including raw material state identification, material difference quantification, and adaptive process control.

[0115] Specifically, the near-infrared reflectance spectral feature vector set can accurately characterize the optical absorption behavior of circuit board surface materials in different bands, reflecting the physical aging degree of raw materials during service, the molecular degradation level of organic resins, the oxidation coverage of metal surfaces, and the enrichment degree of organic residues such as flux or thermal grease, and has the ability to express the service status of materials with high sensitivity.

[0116] The near-infrared reflectance spectral feature vector set was compressed using principal component analysis, retaining the spectral principal component factors that best explain the differences between samples. This resulted in a structured, low-dimensional standard vector form that can be input into machine learning models, significantly improving the input efficiency and stability for subsequent manufacturing age determination, aging level classification, and pyrolysis process adaptability assessment. It also avoided model overfitting, computational redundancy, and feature coupling problems caused by high-dimensional original data.

[0117] In this application, the near-infrared reflectance spectral feature vector set is obtained by principal component analysis and projection processing of the reflectance vectors of multiple circuit boards. Each principal component variable represents the absorption structure variation characteristics of different wavebands and has a clear material response meaning. The following are some representative features related to the aging degree and manufacturing age of the circuit boards:

[0118] The absorption characteristic factor of aged resin, this principal component mainly responds in the 1350 to 1449 nm wavelength range, which belongs to the typical absorption band of functional groups such as CH and C=O in the near-infrared region of epoxy resin matrix (functional groups refer to atomic group structures of organic molecules that exhibit specific properties in chemical reactions). Circuit boards with long service time or severe thermal aging will show a significant decrease in reflectivity in this wavelength range, forming a deep absorption valley. This characteristic factor can be used to characterize the degree of thermo-oxidative aging of the resin matrix.

[0119] The moisture content and hygroscopic residue response factor, primarily loaded in the 1450-1550 nm wavelength range, reflects the absorption behavior of infrared light by -OH groups and physically adsorbed water on the circuit board surface. Raw material boards manufactured earlier and exposed to humid environments for extended periods exhibit higher absorption intensity in this wavelength range; this characteristic helps assess the environmental exposure history of the board during its service life.

[0120] The flux or silicone grease residue identification factor, with its main component concentrated in the 1600-1700 nm region, typically corresponds to a composite absorption structure of organosilicon and fatty acid additives. Some early-process circuit boards, due to the lack of low-residue soldering technology, exhibit significant multi-peak overlap in their reflectivity characteristics in this wavelength range. Therefore, this characteristic can help determine the generational characteristics of the manufacturing process and support the identification of the manufacturing era.

[0121] The surface characteristic factor of the oxide solder layer, a principal component responding in the 1050 to 1250 nm range, reflects the infrared absorption shoulder peak characteristics of the metal surface oxide film (such as SnO2, CuO). Solder layers on circuit boards with longer service lives are typically accompanied by oxidation and darkening, with significant changes in reflectivity in this wavelength band. This principal component variable can be used to infer the solder condition and potential service life.

[0122] It should be noted that the characteristic relationship between wavelength and aging for the representative features illustrated in this application is shown in Table 2, Wavelength-Aging Characteristic Data Table:

[0123] Table 2 Wavelength-Aging Characteristic Data Table

[0124]

[0125] The third processing module is used to perform feature extraction processing on the surface images of the R circuit boards acquired by the image acquisition device to obtain a set of circuit board image feature vectors. The circuit board image feature vectors are used to reflect the appearance and structural state of the circuit board, the distribution of welding processes, and signs of aging.

[0126] For example, the image acquisition device described in this application is preferably an industrial vision camera, employing a linear or area array CMOS or CCD image sensor, and used in conjunction with a standard light source device (such as a diffused LED surface light source or a strip backlight) to acquire image data of the target circuit board in the visible light range (400 to 700 nanometers). The image of each circuit board should cover the entire area, and the image format is preferably an 8-bit grayscale image or an RGB three-channel image.

[0127] The method for obtaining the circuit board image feature vector set includes:

[0128] S200: Let r be initially 1, and the range of r is from 1 to R;

[0129] S201: Preprocess the surface image of the r-th circuit board to obtain a preprocessed surface image of the r-th circuit board; the preprocessing includes brightness normalization, contrast enhancement, edge smoothing and artifact elimination;

[0130] It should be noted that brightness normalization typically employs histogram equalization, redistributing the brightness values ​​of image pixels to make the overall grayscale level of the image more uniform within the output range, thereby eliminating local underexposure or overexposure caused by uneven illumination. Contrast enhancement uses adaptive Gamma correction, dynamically adjusting the Gamma value according to the brightness distribution of the image, giving low-contrast areas a higher dynamic range and thus enhancing image details. For edge smoothing, Gaussian filtering or bilateral filtering algorithms are commonly used. Gaussian filtering smooths the image by constructing a normally distributed convolution kernel, which is suitable for overall noise suppression. Bilateral filtering achieves smoothing while preserving edge structure, and is particularly suitable for scenarios that preserve fine texture features. Image artifact removal often combines morphological reconstruction methods, using a combination of erosion and dilation operations to effectively remove unstructured artifacts caused by background interference, dust residue, or overexposure, maintaining the integrity and continuity of the true structural boundaries in the image.

[0131] This application introduces image preprocessing methods such as brightness normalization, contrast enhancement, edge smoothing, and artifact elimination into the image feature extraction process. The purpose is to improve the image input quality and the accuracy of subsequent feature recognition. These are reasonable optimization choices made to realize the feasibility and adaptability of the technical solution of this invention, and do not constitute the core innovation of the technical solution of this application.

[0132] S202: Divide the preprocessed surface image of the r-th circuit board into H×H equal-area image grids (e.g., 4×4), extract the number of circuit board solder joints in each image grid, and calculate the solder joint distribution uniformity factor of the r-th circuit board; mark the image grids with a number of circuit board solder joints greater than a preset circuit board solder joint number threshold as welding areas, and count the number of welding areas as the number of welding areas; extract the area of ​​the solder joint area and the area of ​​the image grid corresponding to each image grid, and calculate the average solder joint density of the r-th circuit board; for example, in this application, the circuit board solder joint number threshold can be set to 10.

[0133] S203: Construct the circuit board image feature vector of the r-th circuit board from the solder joint distribution uniformity factor, the number of solder areas and the average solder joint density, and add it to the circuit board image feature vector set.

[0134] S204: Let r = r + 1. If r is less than or equal to R, return to S201 and continue execution. If r is greater than R, end the current process.

[0135] The calculation method for the weld joint distribution uniformity factor includes:

[0136] ;

[0137] in, This is the solder joint distribution uniformity factor. This represents the number of solder joints on the circuit board in the h-th image grid within an H×H equal-area image grid. This represents the average number of solder joints on the circuit board. The solder joint distribution uniformity factor represents the standard deviation of the number of solder joints on a circuit board. It is calculated by dividing the standard deviation by the mean number of solder joints. When solder joints are evenly distributed across all areas, the standard deviation is small, and the solder joint distribution uniformity factor approaches 0. Conversely, if there are concentrated or large blank areas, the standard deviation increases, and the solder joint distribution uniformity factor approaches 1. The solder joint distribution uniformity factor is used to measure whether there are excessively dense or sparse areas without solder joints in the solder joint layout of a circuit board, thus reflecting its process complexity and alloy distribution strategy.

[0138] The methods for calculating the average solder joint density of the r-th circuit board include:

[0139] ;

[0140] in, Let h be the average solder joint density of the r-th circuit board, and h be the index variable in the summation formula. Let H be the area of ​​the solder joint region corresponding to the h-th image grid in an H×H equal-area image grid. Let h be the area of ​​the h-th image grid in a H×H grid of equal-area images.

[0141] It should be noted that, in this embodiment of the invention, the constructed circuit board image feature vector set extracts observable physical features of the circuit board's soldering structure at the image level, thereby achieving indirect modeling and quantifiable evaluation of the circuit board's manufacturing age and aging status. The circuit board image feature vector includes solder joint distribution uniformity factor, number of soldering areas, and average solder joint density, exhibiting statistical behavior patterns highly correlated with process iteration level, manufacturing batch characteristics, and long-term service status, and can serve as an auxiliary basis for identifying the manufacturing age and degree of aging.

[0142] Specifically, circuit boards manufactured in different eras typically employ different generations of soldering process standards and layout templates. For instance, early circuit boards, due to a higher proportion of manual or semi-automatic soldering, often exhibit uneven solder joint distribution, localized over-soldering, or inconsistent pad sizes. Newer generation circuit boards, on the other hand, mostly utilize precision surface mount technology and fully automated soldering, resulting in more regular solder joint distribution and higher density. Therefore, by analyzing the solder joint distribution uniformity factor and the number of soldered areas, the possible manufacturing process era of a circuit board can be determined in a passive state.

[0143] On the other hand, after long-term operation, the morphology of the solder joints of the circuit board may also undergo slight degradation due to factors such as thermal expansion and contraction, current thermal stress and environmental pollution. For example, the shape of the solder reflow may become blurred, the corrosion distortion of the pad boundary may occur, or the area of ​​the soldering region may grow unstablely. These degradation phenomena can be manifested in the image solder joint density statistics and regional connectivity as abnormal solder joint size, blurred edges or abnormal fluctuations in the number of soldering regions, thus providing quantitative clues at the image level for aging condition identification.

[0144] In summary, by constructing a unified image feature data set from multiple circuit board image feature parameters and combining it with spectral data and operating condition parameters to achieve fusion modeling, the system's ability to identify manufacturing age and aging state is enhanced. It also provides key visual structural input for the subsequent adaptive adjustment model of smelting and pyrolysis parameters, thereby improving the system's adaptability and intelligence level.

[0145] It should be noted that the flowchart of the feature extraction method for the R-block circuit board is as follows: Figure 3 As shown, Figure 3This includes methods for obtaining sets of fluorescence spectral feature vectors, near-infrared reflectance spectral feature vectors, and circuit board image feature vectors, ultimately constructing a feature dataset.

[0146] The comprehensive evaluation module evaluates R circuit boards based on a set of fluorescence spectral feature vectors, a set of near-infrared reflectance spectral feature vectors, and a set of circuit board image feature vectors, obtaining the predicted manufacturing age and aging assessment level of the R circuit boards. The predicted manufacturing age refers to the preset time interval to which the circuit board was manufactured, such as "before 2010", "2010 to 2015", "2016 to 2020", and "after 2021".

[0147] Methods for obtaining the predicted manufacturing age and aging assessment level of R-block circuit boards include:

[0148] S300: Let r be initially 1, and the range of r is from 1 to R;

[0149] S301: Obtain the fluorescence spectrum feature vector of the r-th circuit board from the fluorescence spectrum feature vector set, obtain the near-infrared reflectance spectrum feature vector of the r-th circuit board from the near-infrared reflectance spectrum feature vector set, and obtain the circuit board image feature vector of the r-th circuit board from the circuit board image feature vector set.

[0150] S302: Input the fluorescence spectral feature vector, near-infrared reflectance spectral feature vector and circuit board image feature vector of the r-th circuit board into the comprehensive evaluation model to obtain the predicted manufacturing age and aging assessment level of the r-th circuit board.

[0151] S303: Let r = r + 1. If r is less than or equal to R, return to S301 to continue execution. If r is greater than R, end the current process.

[0152] The training method for the comprehensive evaluation model includes:

[0153] A comprehensive evaluation dataset is pre-constructed, which includes comprehensive evaluation data of the ZH group and the predicted manufacturing year and aging evaluation level corresponding to the comprehensive evaluation data of the ZH group, where ZH is a positive integer; the comprehensive evaluation data includes fluorescence spectral feature vectors, near-infrared reflectance spectral feature vectors, and circuit board image feature vectors; the comprehensive evaluation dataset is divided into a training set and a validation set, the training set is used for learning the parameters of the comprehensive evaluation model, and the validation set is used to monitor the generalization performance and overfitting degree of the comprehensive evaluation model in real time;

[0154] A deep neural network with a multilayer perceptron as its core is used as the comprehensive evaluation model. The comprehensive evaluation data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer uses a nonlinear activation function to extract high-order features, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each predicted manufacturing age and aging assessment level. Finally, the predicted manufacturing age and aging assessment level corresponding to the highest probability are taken as the prediction results of the comprehensive evaluation model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is set: when the prediction accuracy on the validation set reaches or exceeds a preset threshold, the comprehensive evaluation model is determined to have converged and training is terminated.

[0155] In this embodiment of the invention, by constructing a comprehensive evaluation model and inputting the multi-source feature vectors (including fluorescence spectral feature vectors, near-infrared reflectance spectral feature vectors, and circuit board image feature vectors) corresponding to each circuit board into the comprehensive evaluation model for joint analysis and inference processing, the predicted manufacturing age and aging assessment level of each circuit board can be effectively obtained. This process not only achieves high-precision non-destructive identification of the service background and aging state of the circuit board, but also significantly improves the pertinence and adaptability of subsequent melting and recycling parameter settings. The comprehensive evaluation model has cross-domain feature fusion capabilities and strong generalization recognition capabilities, and is particularly suitable for circuit board reuse scenarios where the proportion of old samples is small and material changes are complex.

[0156] Specifically, fluorescence spectral feature vectors reflect the composition, content distribution, and ratio changes of metallic elements on the circuit board surface, indirectly mapping the differences in solder, metal coatings, and other material systems used in different generations of circuit boards; near-infrared reflectance spectral feature vectors reveal the decay characteristics of molecular functional groups in the circuit board's dielectric layer, polymer materials, or encapsulation coatings, possessing strong indicative significance for material aging; and circuit board image feature vectors capture visualized manufacturing and degradation information such as the circuit board's structural layout, solder joint density, and distribution uniformity. These three types of feature vectors together constitute a feature fusion space with representational, complementary, and dimensionality-reduction abstraction capabilities.

[0157] Compared with existing technologies, this application does not rely on traditional serial number parsing, RFID chip tags, or physical batch documents. Instead, it uses a data-driven soft inference method to accurately identify the manufacturing background and service degradation status of tagged circuit boards. This significantly improves the intelligent sensing and adaptation capabilities of automated recycling and smelting pyrolysis processes, effectively avoiding problems such as low processing efficiency, excessive energy consumption, or increased pollution risks caused by misjudgment of manufacturing batches or misestimation of aging levels. It has outstanding practical value and promising prospects for industrial application.

[0158] The parameter control module generates pyrolysis parameter adjustment instructions based on the set of fluorescence spectral feature vectors, the set of near-infrared reflectance spectral feature vectors, the set of circuit board image feature vectors, the predicted manufacturing age and aging assessment level of R circuit boards, and performs intelligent adaptive control of the circuit board pyrolysis parameters.

[0159] Methods for intelligent adaptive control of circuit board pyrolysis parameters include:

[0160] Based on the predicted manufacturing age and aging assessment level of R-block circuit boards, a set of binary pairs of predicted manufacturing age and a set of binary pairs of aging assessment level are constructed.

[0161] The dominant component was determined and the dominant label value was constructed based on the binary set of predicted manufacturing age and the binary set of aging assessment level.

[0162] The dominant label value, fluorescence spectral feature vector set, near-infrared reflectance spectral feature vector set, and circuit board image feature vector set are input into the pyrolysis parameter setting model to obtain the corresponding pyrolysis parameter combination. The pyrolysis parameter combination includes the target pyrolysis temperature, heating rate, pyrolysis residence time, oxygen to inert gas ratio, and initial pressure of the pyrolysis reaction chamber.

[0163] The system control platform sends the pyrolysis parameter combination to the corresponding control equipment and executes it.

[0164] The methods for obtaining the predicted manufacturing age binary set and the aging assessment level binary set include:

[0165] S400: Let r be initially 1, and the range of r is from 1 to R;

[0166] S401: Obtain the predicted manufacturing age and aging assessment level of the r-th circuit board;

[0167] Determine whether the predicted manufacturing year already exists in the predicted manufacturing year tuple set; if it exists, increment the circuit board count value of the predicted manufacturing year tuple by 1; if it does not exist, construct a predicted manufacturing year tuple from the predicted manufacturing year of the r-th circuit board and 1, and add it to the predicted manufacturing year tuple set; an example of the predicted manufacturing year tuple is (predicted manufacturing year of the r-th circuit board, circuit board count value).

[0168] Determine whether the aging assessment level already exists in the aging assessment level binary set; if it exists, increment the circuit board count value of the aging assessment level binary set by 1; if it does not exist, construct an aging assessment level binary set by the aging assessment level of the r-th circuit board and 1, and add it to the aging assessment level binary set; an example of the aging assessment level binary set is (aging assessment level of the r-th circuit board, circuit board count value).

[0169] S402: Let r = r + 1. If r is less than or equal to R, return to S401 and continue execution; if r is greater than R, end the current process.

[0170] like Figure 4 As shown, the method for determining the dominant component and constructing the dominant label value based on the set of predicted manufacturing year binary data and the set of aging assessment level binary data includes:

[0171] Initialize the dominant identifier for predicting the manufacturing year and the dominant identifier for the aging assessment level to No;

[0172] The circuit board quantity count value of each predicted manufacturing year binary set is divided with R to obtain the corresponding circuit board quantity percentage, and the first quantity percentage set is constructed.

[0173] The circuit board quantity count value of each aging assessment level binary set in the aging assessment level binary set is divided with R to obtain the corresponding circuit board quantity percentage, and a second quantity percentage set is constructed.

[0174] Select the largest first quantity percentage from the first quantity percentage set, and determine whether the first quantity percentage is greater than the preset first quantity percentage threshold. If the determination result is yes, then set the dominant identifier of the predicted manufacturing year to yes.

[0175] Select the largest second quantity percentage from the second quantity percentage set, and determine whether the second quantity percentage is greater than the preset second quantity percentage threshold. If the determination result is yes, then set the dominant indicator of the aging assessment level to yes.

[0176] For example, the first quantity percentage threshold and the second quantity percentage threshold in this application can be set to 60%.

[0177] The dominant identifiers for predicting the manufacturing age and aging assessment level are matched with a pre-constructed dominant label value mapping table to obtain the dominant label value; the dominant label value mapping table includes the dominant identifiers for predicting the manufacturing age, the dominant identifiers for aging assessment level, and the corresponding dominant label values.

[0178] An example of the dominant label value mapping table is shown in Table 3:

[0179] Table 3. Dominant Label Numerical Mapping Table

[0180]

[0181] It should be noted that, in this embodiment of the invention, the dominant identifiers of the predicted manufacturing year and the aging assessment level are identified by a statistical distribution constructed based on the set of binary pairs of predicted manufacturing year and the set of binary pairs of aging assessment level. These are then matched with a dominant label value mapping table to obtain the dominant label value. This dominant label value serves as a unified control strategy index identifier, simplifying the subsequent pyrolysis parameter setting logic and improving the efficiency and reusability of the decision-making process.

[0182] Specifically, the dominant label value enables rapid indexing of four combined states through numerical encoding, facilitating its use and expansion in control models, parameter configuration tables, or logical decision nodes. By introducing the dominant label value, a label-driven parameter control strategy can be constructed, enabling rapid parameter matching for different types of circuit board batches. For example, when the dominant label value is 1, indicating that both the predicted manufacturing year and aging level are dominant, high-confidence parameters can be prioritized. When the dominant label value is 4, indicating that neither attribute is dominant, the fusion regression model is triggered to dynamically calculate the parameter set, achieving refined adaptive control and thus improving the system's classification response capability for batches with different feature distributions.

[0183] The training method for the pyrolysis parameter setting model includes:

[0184] A pyrolysis parameter setting dataset is pre-constructed, comprising QH group pyrolysis parameter setting data and corresponding combinations of pyrolysis parameters, where QH is a positive integer. The pyrolysis parameter setting data includes dominant label values, a set of fluorescence spectral feature vectors, a set of near-infrared reflectance spectral feature vectors, and a set of circuit board image feature vectors. The pyrolysis parameter setting dataset is divided into a training set and a validation set. The training set is used for learning the parameters of the pyrolysis parameter setting model, and the validation set is used to monitor the generalization performance and overfitting degree of the pyrolysis parameter setting model in real time.

[0185] A deep neural network with a multilayer perceptron as its core is used as the pyrolysis parameter setting model. The pyrolysis parameter setting data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer uses a nonlinear activation function to extract high-order features, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each combination of pyrolysis parameters. Finally, the combination of pyrolysis parameters with the highest probability is taken as the prediction result of the pyrolysis parameter setting model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is set: when the prediction accuracy on the validation set reaches or exceeds a preset threshold, the pyrolysis parameter setting model is determined to have converged and training is terminated.

[0186] To further clarify the application of the pyrolysis parameter settings in this application within the actual pyrolysis process, the following example, using a specific control scenario, illustrates the parameter execution and adaptive adjustment process driven by the dominant label value. A schematic diagram of the data-driven input and control linkage of the circuit board pyrolysis parameters is shown below. Figure 5 As shown.

[0187] Assuming the current processing object is a batch of R=120 discarded circuit boards, the system has obtained the fluorescence spectral feature vector, near-infrared reflectance spectral feature vector and image feature vector of each circuit board through the first processing module, the second processing module and the third processing module, and constructed a binary set of predicted manufacturing age and a binary set of aging assessment level.

[0188] Based on the proportions, the system identified 78 circuit boards in this batch with a predicted manufacturing year of "2015–2017," accounting for approximately 65%, exceeding the first quantity percentage threshold of 60%; and 74 circuit boards with an aging assessment level of "Level 3," accounting for approximately 61.6%, also exceeding the second quantity percentage threshold of 60%. Therefore, the system determined that both the dominant predicted manufacturing year and the dominant aging assessment level were valid, and based on the dominant label value mapping table, the dominant label value was determined to be "1."

[0189] Subsequently, the system inputs the dominant label value, fluorescence spectral feature vector set, near-infrared reflectance spectral feature vector set, and circuit board image feature vector set into the pyrolysis parameter setting model, and outputs the following pyrolysis parameter combination:

[0190] Target pyrolysis temperature = 820°C;

[0191] Heating rate = 6.5°C / min;

[0192] Pyrolysis residence time = 22 min;

[0193] The ratio of oxygen to inert gas is 30:70.

[0194] The initial pressure in the pyrolysis reaction chamber is 0.85 atm.

[0195] The system automatically sends the above parameter commands to the pyrolysis control equipment: setting the target pyrolysis temperature to the main heating device temperature control equipment; adjusting the heating rate to the heating speed controller; configuring the pyrolysis residence time to the time controller; configuring the oxygen and inert gas ratio injection atmosphere mixing valve group controller; and configuring the initial pressure of the pyrolysis reaction chamber to the inlet pressure regulating unit. Ultimately, under various mixed fluctuation conditions of circuit boards, the risk of coking and dioxin exceeding standards can be significantly reduced, and the energy efficiency, safety, and intelligence level of circuit board resource pyrolysis treatment can be significantly improved.

[0196] Example 2:

[0197] Please see Figure 2 As shown, this embodiment provides a multimodal sensing adaptive control method for circuit board pyrolysis parameters, including:

[0198] The spectral signals of the R-block circuit board acquired in the fluorescence spectral acquisition area are analyzed and their features are extracted to obtain a set of fluorescence spectral feature vectors;

[0199] Feature extraction and structuring are performed on the reflectance spectral data of the R-block circuit board collected in the near-infrared reflectance spectral acquisition area to obtain a set of near-infrared reflectance spectral feature vectors;

[0200] Feature extraction processing is performed on the surface images of R circuit boards acquired by the image acquisition device to obtain a set of circuit board image feature vectors;

[0201] The R-block circuit boards are evaluated based on the fluorescence spectral feature vector set, the near-infrared reflectance spectral feature vector set, and the circuit board image feature vector set to obtain the predicted manufacturing age and aging assessment level of the R-block circuit boards.

[0202] Based on the set of fluorescence spectral feature vectors, the set of near-infrared reflectance spectral feature vectors, the set of circuit board image feature vectors, the predicted manufacturing age and aging assessment level of R circuit boards, pyrolysis parameter adjustment instructions are generated, and intelligent adaptive control of the circuit board pyrolysis parameters is performed.

[0203] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0204] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multimodal sensing adaptive control method for circuit board pyrolysis parameters, characterized in that, include: The spectral signals of the R-block circuit board acquired in the fluorescence spectral acquisition area are analyzed and their features are extracted to obtain a set of fluorescence spectral feature vectors; Feature extraction and structuring are performed on the reflectance spectral data of the R-block circuit board collected in the near-infrared reflectance spectral acquisition area to obtain a set of near-infrared reflectance spectral feature vectors; Feature extraction processing is performed on the surface images of R circuit boards acquired by the image acquisition device to obtain a set of circuit board image feature vectors; The R-block circuit boards are evaluated based on the fluorescence spectral feature vector set, the near-infrared reflectance spectral feature vector set, and the circuit board image feature vector set to obtain the predicted manufacturing age and aging assessment level of the R-block circuit boards. Based on the set of fluorescence spectral feature vectors, the set of near-infrared reflectance spectral feature vectors, the set of circuit board image feature vectors, the predicted manufacturing age and aging assessment level of R circuit boards, pyrolysis parameter adjustment instructions are generated, and intelligent adaptive control of the circuit board pyrolysis parameters is performed. Methods for intelligent adaptive control of circuit board pyrolysis parameters include: Based on the predicted manufacturing age and aging assessment level of R-block circuit boards, a set of binary pairs of predicted manufacturing age and a set of binary pairs of aging assessment level are constructed. The dominant component was determined and the dominant label value was constructed based on the binary set of predicted manufacturing age and the binary set of aging assessment level. The dominant label value, fluorescence spectral feature vector set, near-infrared reflectance spectral feature vector set, and circuit board image feature vector set are input into the pyrolysis parameter setting model to obtain the corresponding pyrolysis parameter combination. The pyrolysis parameter combination includes the target pyrolysis temperature, heating rate, pyrolysis residence time, oxygen to inert gas ratio, and initial pressure of the pyrolysis reaction chamber. The system control platform sends the pyrolysis parameter combination to the corresponding control equipment and executes it.

2. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 1, characterized in that, The methods for obtaining the predicted manufacturing age binary set and the aging assessment level binary set include: S400: Let r be initially 1, and the range of r is from 1 to R; S401: Obtain the predicted manufacturing age and aging assessment level of the r-th circuit board; Determine whether the predicted manufacturing year already exists in the predicted manufacturing year set; if it exists, increment the circuit board count value of the predicted manufacturing year set by 1; if it does not exist, construct the predicted manufacturing year set of the r-th circuit board and 1 into a predicted manufacturing year set, and add it to the predicted manufacturing year set. Determine whether the aging assessment level already exists in the aging assessment level binary set; if it exists, increment the circuit board count value of the aging assessment level binary set of that aging assessment level by 1; if it does not exist, construct an aging assessment level binary set by combining the aging assessment level of the r-th circuit board with 1, and add it to the aging assessment level binary set. S402: Let r = r + 1. If r is less than or equal to R, return to S401 and continue execution; if r is greater than R, end the current process.

3. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 1, characterized in that, Methods for determining the dominant component and constructing the dominant label value based on the binary set of predicted manufacturing age and the binary set of aging assessment level include: Initialize the dominant identifier for predicting the manufacturing year and the dominant identifier for the aging assessment level to No; The circuit board quantity count value of each predicted manufacturing year binary set is divided with R to obtain the corresponding circuit board quantity percentage, and the first quantity percentage set is constructed. The circuit board quantity count value of each aging assessment level binary set in the aging assessment level binary set is divided with R to obtain the corresponding circuit board quantity percentage, and a second quantity percentage set is constructed. Select the largest first quantity percentage from the first quantity percentage set, and determine whether the first quantity percentage is greater than the preset first quantity percentage threshold. If the determination result is yes, then set the dominant identifier of the predicted manufacturing year to yes. Select the largest second quantity percentage from the second quantity percentage set, and determine whether the second quantity percentage is greater than the preset second quantity percentage threshold. If the determination result is yes, then set the dominant indicator of the aging assessment level to yes. The dominant identifiers for predicting the manufacturing age and aging assessment level are matched with a pre-constructed dominant label value mapping table to obtain the dominant label values.

4. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 1, characterized in that, Methods for obtaining the predicted manufacturing age and aging assessment level of R-block circuit boards include: Based on the set of fluorescence spectral feature vectors, the set of near-infrared reflectance spectral feature vectors, and the set of circuit board image feature vectors, multimodal feature vectors for each circuit board are obtained, and the multimodal feature vectors are input into the comprehensive evaluation model to obtain the predicted manufacturing age and aging assessment level of each circuit board; the multimodal feature vectors are fluorescence spectral feature vectors, near-infrared reflectance spectral feature vectors, and circuit board image feature vectors.

5. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 1, characterized in that, The method for obtaining the set of fluorescence spectral feature vectors includes: The X-ray fluorescence spectroscopy acquisition device emits a micro-focus X-ray beam with a focal spot size of less than 100 micrometers onto the surface of the R-block circuit board; under the bombardment of X-rays, the inner-shell electrons of the elements on the surface of the circuit board undergo transitions and instantaneously release X-ray fluorescence with characteristic energy; After X-ray fluorescence enters the spectrometer detector, each fluorescent photon of the X-ray fluorescence forms an independent incident photon event. The spectrometer detector collects and converts the electronic signals released by the incident photon events, generating a pulsed voltage signal whose amplitude is linearly related to the fluorescence energy. After being pre-amplified, shaped and converted from analog to digital, the pulsed voltage signal enters the multichannel analyzer for energy channel classification and statistical accumulation. The spectral analysis software automatically extracts the peak energy, peak height and relative ratio of each element in the R-block circuit board. The peak energy, peak intensity, and relative ratio of the R circuit boards are normalized and vectorized to construct R sets of fluorescence spectral feature vectors, which are then used to construct a set of fluorescence spectral feature vectors.

6. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 1, characterized in that, The method for obtaining the near-infrared reflectance spectral feature vector set includes: The near-infrared reflectance spectrometer emits near-infrared light with a wavelength range of 900 to 1700 nanometers onto the surface of R circuit boards. The near-infrared detector receives the near-infrared light signal reflected back from the surface of the circuit boards and samples it according to the number of wavelength points Q. That is, the reflectance of each circuit board is collected at Q wavelength points to form a Q-dimensional reflectance vector. Principal component analysis is performed on the Q-dimensional reflectance vector corresponding to each circuit board to obtain the near-infrared reflectance spectral feature vectors corresponding to R circuit boards; the near-infrared reflectance spectral feature vectors corresponding to R circuit boards are then used to construct a set of near-infrared reflectance spectral feature vectors.

7. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 6, characterized in that, The methods for obtaining the near-infrared reflectance spectral feature vector corresponding to the R-block circuit board include: Construct a reflectivity matrix from the Q-dimensional reflectivity vector corresponding to each circuit board; Subtract the mean reflectance of each column from the reflectance data in the reflectance matrix to obtain a zero-mean matrix. Calculate the covariance matrix based on the zero-mean matrix. The covariance matrix is ​​used to measure the linear correlation between reflectance values ​​of different bands. Find the eigenvalues ​​of the covariance matrix to obtain W eigenvalues. Find the corresponding eigenvector for each eigenvalue. Sort the W eigenvalues ​​in descending order to form an eigenvalue set. Then, construct an eigenvector set from the eigenvalue set and the corresponding eigenvectors of the eigenvalue set. The number of principal components (ZCF) is determined based on the eigenvalue set and a preset cumulative contribution rate threshold. The first ZCF eigenvectors are selected from the eigenvector set to construct the principal component vector set; Perform linear projection operations on the Q-dimensional reflectance vector of the R-block circuit board with the principal component vector set to obtain the principal component projection value sequence corresponding to the R-block circuit board. The principal component projection value sequence is the near-infrared reflectance spectral feature vector of the circuit board.

8. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 7, characterized in that, Methods for determining the number of principal components (ZCF) based on the eigenvalue set and a preset cumulative contribution rate threshold include: Step 1: Sum the eigenvalues ​​in the eigenvalue set to obtain the total eigenvalues; set the initial value of the principal component number ZCF to 1; Step 2: Select the first ZCF eigenvalues ​​from the eigenvalue set and sum them to obtain the cumulative eigenvalue. Divide the cumulative eigenvalue by the sum of the eigenvalues ​​to obtain the cumulative contribution rate. Step 3: If the cumulative contribution rate is greater than the cumulative contribution rate threshold, the number of principal components ZCF is obtained, and the current process ends; if the cumulative contribution rate is less than or equal to the cumulative contribution rate threshold, ZCF = ZCF + 1 is set, and the process returns to step 2 to continue execution.

9. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 7, characterized in that, The methods for obtaining the principal component projection value sequence corresponding to the R-block circuit board include: S100: Construct a principal component matrix from the set of principal component vectors; the principal component matrix has Q rows and ZCF columns; each column in the principal component matrix corresponds to a principal component vector; let the initial value of r be 1, and the value of r range from 1 to R; S101: Obtain the element of the r-th row of the zero-mean matrix, denoted as the centered reflectivity vector; S102: Multiply the centered reflectivity vector by the principal component matrix to obtain the sequence of principal component projection values; S103: Let r = r + 1. If r is less than or equal to R, return to S101 and continue execution. If r is greater than R, end the current process.

10. The multimodal sensing adaptive control method for circuit board pyrolysis parameters according to claim 1, characterized in that, The method for obtaining the circuit board image feature vector set includes: S200: Let r be initially 1, and the range of r is from 1 to R; S201: Preprocess the surface image of the r-th circuit board to obtain a preprocessed surface image of the r-th circuit board; the preprocessing includes brightness normalization, contrast enhancement, edge smoothing and artifact elimination; S202: Divide the preprocessed surface image of the r-th circuit board into H×H equal-area image grids, extract the number of circuit board solder joints in each image grid, and calculate the solder joint distribution uniformity factor of the r-th circuit board; mark the image grids with a number of circuit board solder joints greater than a preset threshold as welding areas, and count the number of welding areas as the number of welding areas; extract the area of ​​the solder joint area and the area of ​​the image grid corresponding to each image grid, and calculate the average solder joint density of the r-th circuit board; S203: Construct the circuit board image feature vector of the r-th circuit board from the solder joint distribution uniformity factor, the number of solder areas and the average solder joint density, and add it to the circuit board image feature vector set. S204: Let r = r + 1. If r is less than or equal to R, return to S201 and continue execution. If r is greater than R, end the current process.

11. A multimodal sensing adaptive control system for pyrolysis parameters of a circuit board, used to implement the multimodal sensing adaptive control method for pyrolysis parameters of a circuit board as described in any one of claims 1-10, characterized in that, include: The first processing module is used to analyze and extract features from the spectral signals of the R-block circuit board acquired in the fluorescence spectral acquisition area to obtain a set of fluorescence spectral feature vectors. The second processing module is used to extract features and perform structured processing on the reflectance spectral data of the R-block circuit board collected in the near-infrared reflectance spectral acquisition area to obtain a set of near-infrared reflectance spectral feature vectors. The third processing module is used to perform feature extraction processing on the surface image of the R-block circuit board acquired by the image acquisition device to obtain a set of circuit board image feature vectors. The comprehensive evaluation module evaluates the R circuit boards based on the fluorescence spectral feature vector set, the near-infrared reflectance spectral feature vector set, and the circuit board image feature vector set, and obtains the predicted manufacturing age and aging assessment level of the R circuit boards. The parameter control module generates pyrolysis parameter adjustment instructions based on the set of fluorescence spectral feature vectors, the set of near-infrared reflectance spectral feature vectors, the set of circuit board image feature vectors, the predicted manufacturing age and aging assessment level of R circuit boards, and performs intelligent adaptive control of the circuit board pyrolysis parameters.

Images