Notebook computer shell injection molding part surface quality detection method and system

Abstract

The invention relates to the technical field of optical detection, in particular to a notebook computer shell injection molding part surface quality detection method and system, and the method comprises the steps: irradiating a standard part with a preset wave band according to preset parameters through an infrared spectrometer to obtain a first spectrogram, and irradiating the injection molding part with the preset wave band according to the preset parameters to obtain a second spectrogram. Through the synergistic effect of a non-negative matrix factorization algorithm and a Transform fusion model, the problem of spectrum overlapping caused by addition of a flame retardant is effectively solved, through seven-degree polynomial and smooth second derivative processing, the carbonyl peak position measurement error is further reduced, and compared with a traditional Fourier transform infrared spectroscopy, the feature parameter extraction precision is further improved. The number of the carbonyl absorption peaks and the positions of the carbonyl absorption peaks are compared with the standard threshold values twice, so that the problem that some qualified products are mistakenly considered as non-qualified products easily due to deviation during judgment of the carbonyl content of the injection molding shell of the notebook computer is further solved.

Description

Technical Field

[0001] The present invention relates to the technical field of optical detection, in particular to a method and system for detecting the surface quality of injection-molded parts of a notebook shell. Background Art

[0002] PC / ABS alloy, an engineering plastic with excellent mechanical and processing properties, is widely used in the manufacture of laptop computer cases through the composite modification of polycarbonate (PC) and acrylonitrile-butadiene-styrene copolymer (ABS). However, in actual large-scale production, the mixing of old and new equipment presents significant quality risks. Due to the inaccurate temperature control system (errors of up to ±10°C), high barrel temperatures (typically reaching 250-300°C) and mold thermal cycling in older equipment, thermal oxidative degradation of the PC / ABS alloy is exacerbated, leading to molecular chain breakage and the formation of numerous carbonyl (C=O) structures. The presence of these structures not only degrades the material's mechanical properties but also causes cosmetic defects such as yellowing and gloss loss on the housing surface. These defects particularly affect the color uniformity of subsequent surface treatments such as spray painting and silk screen printing, and can even lead to rework and re-injection of batches due to reduced coating adhesion. Therefore, accurate identification of carbonyl content is a critical step in quality control. However, to meet stringent laptop fire safety standards (such as the UL94V-0 flame retardancy rating), 10%-20% by mass of decabromodiphenylethane flame retardant must be added to the raw materials. The addition of this flame retardant causes interference peaks in the infrared spectrum band of 1650-1750 cm⁻¹, generated by the polybrominated aromatic hydrocarbon groups in its molecular structure, to overlap with the carbonyl absorption peak (1680-1750 cm⁻¹). Furthermore, the matrix absorption enhancement effect caused by high addition levels significantly increases spectral background noise, severely interfering with the identification of carbonyl signals. Existing detection technologies are limited by flame retardant interference and spectral signal overlap, leading to inaccuracies in the determination of carbonyl content in injection-molded laptop cases. This can easily result in some qualified products being mistaken for substandard ones, leading to production losses for companies. Summary of the Invention

[0003] (1) Technical problems to be solved The purpose of the present invention is to provide a method and system for detecting the surface quality of injection-molded parts of laptop shells, so as to solve the problem of inaccurate judgment of carbonyl content due to interference from flame retardants and overlap of spectral signals.

[0004] (2) Technical solution To achieve the above objectives, the present invention provides a method for detecting the surface quality of an injection-molded part of a notebook housing, comprising: Using an infrared spectrometer, the standard part is irradiated with a preset band according to preset parameters to obtain a first spectrum graph, and the injection molded part is irradiated with a preset band according to preset parameters to obtain a second spectrum graph; the first spectrum graph and the second spectrum graph are Fourier transformed to obtain first spectrum data and second spectrum data respectively.

[0005] The first spectral data and the second spectral data are preprocessed with baseline drift correction and intensity normalization, respectively; the preprocessed first spectral data and the second spectral data are feature extracted by polynomial fitting and derivative spectral analysis to obtain first characteristic spectral data and second characteristic spectral data, respectively; the first characteristic spectral data include carbonyl absorption peak parameters, and the second characteristic spectral data include absorption peak parameters of the flame retardant and the carbonyl group, absorption peak positions of the flame retardant and the carbonyl group, and the number of absorption peaks of the flame retardant and the carbonyl group.

[0006] The first characteristic spectrum data and the second characteristic spectrum data are input into the Transformer fusion model based on the attention mechanism; the first characteristic spectrum data is decomposed and coupled analyzed to obtain the characteristic parameters of the carbonyl absorption peak; the flame retardant absorption peak parameters and the carbonyl absorption peak parameters in the second characteristic spectrum data are separated by bringing the characteristic parameters of the carbonyl absorption peak and the second characteristic spectrum data into the non-negative matrix decomposition model to obtain the carbonyl absorption peak position and the number of carbonyl absorption peaks; based on the comparison of the carbonyl absorption peak position and the number of carbonyl absorption peaks with the standard threshold, a qualified or unqualified result is output.

[0007] Furthermore, the method of irradiating the injection molded part with an infrared spectrometer at a preset wavelength and according to preset parameters to obtain a first spectrum, and irradiating the injection molded part with a preset wavelength and according to preset parameters to obtain a second spectrum; and obtaining first spectral data and second spectral data by Fourier transforming the first spectrum and the second spectrum includes: The first spectrum was obtained by irradiating the standard sample multiple times with an infrared spectrometer using the 1650-1750 cm⁻¹ band and a scanning resolution of 2-4 cm⁻¹.

[0008] The second spectrum is obtained by irradiating the sample to be tested multiple times with an infrared spectrometer using the 1650-1750 cm⁻¹ band and a scanning resolution of 2-4 cm⁻¹.

[0009] Fourier transform is performed on the first spectrum image and the second spectrum image respectively, and the time domain interference spectrum image is converted into frequency domain spectrum data to obtain first spectrum data and second spectrum data.

[0010] Furthermore, the method of performing baseline drift correction and intensity normalization preprocessing on the first spectral data and the second spectral data, respectively; extracting characteristic parameters from the preprocessed first spectral graph and the second spectral graph by polynomial fitting and derivative spectral analysis, respectively, to obtain the first characteristic spectral data and the second characteristic spectral data includes: The ALS algorithm is used to perform baseline correction on the first and second spectral data respectively, and the parameter λ is set to 10 5 , p=0.01; the spectral intensities of the first spectral data and the second spectral data are mapped to the [0,1] interval through Min-Max normalization.

[0011] For the preprocessed first spectral data and second spectral data, the 1650-1750 cm⁻¹ band is defined as the analysis range, wherein the 1700-1740 cm⁻¹ band is preset as the carbonyl absorption peak region, and the 1650-1680 cm⁻¹ band is preset as the flame retardant absorption peak region.

[0012] The spectral curves in the first and second spectral data were fitted using a seventh-order polynomial. The fitting interval covered 50 cm⁻¹ on both sides of the target peak. The coefficients were optimized using the least squares method to make the sum of squares of the fitting residuals less than or equal to 0.005 AU².

[0013] The fitted first spectrum data and the second spectrum data are respectively smoothed by second-order derivative processing to separate overlapping peaks to obtain first characteristic spectrum data and second characteristic spectrum data.

[0014] Furthermore, the method of inputting the first characteristic spectrum data and the second characteristic spectrum data into a Transformer fusion model based on the attention mechanism; and decomposing and coupling the first characteristic spectrum data to obtain characteristic parameters of the carbonyl absorption peak includes: The first feature spectrum data and the second feature spectrum data are input into a 12-layer Transformer encoder architecture, where each layer contains 8 attention heads and a feedforward neural network.

[0015] Linear projection is performed on the input first characteristic spectrum data and the second characteristic spectrum data and sinusoidal position encoding is added to establish temporal correlation.

[0016] The attention weight is calculated through the Query, Key, and Value matrices. The first feature spectrum data is used as the Query and the second feature spectrum data as the Key / Value to calculate the cross-sample feature correlation and output the first fused feature vector.

[0017] Based on the first fused eigenvector, the first characteristic spectrum data is decomposed and coupled analyzed to obtain characteristic parameters of the carbonyl absorption peak.

[0018] Furthermore, the method of separating the flame retardant absorption peak parameters and the carbonyl absorption peak parameters in the second characteristic spectrum data by substituting the characteristic parameters of the carbonyl absorption peak and the second characteristic spectrum data into a non-negative matrix decomposition model to obtain the carbonyl absorption peak position and the number of carbonyl absorption peaks; and outputting a qualified or unqualified result based on the comparison of the number of carbonyl absorption peaks and the carbonyl absorption peak position with the standard threshold value includes: The characteristic parameters of the carbonyl absorption peak are standardized; a non-negative matrix decomposition model is constructed based on the second characteristic spectrum data; the non-negative matrix decomposition model is the product of the first matrix and the second matrix, where the first matrix is ​​the characteristic parameters of the carbonyl absorption peak, and the second matrix is ​​the absorption peak parameters of the flame retardant and the carbonyl.

[0019] The non-negative matrix factorization model is trained through an iterative optimization algorithm, and the element values ​​in the first matrix and the second matrix are adjusted until convergence; the carbonyl absorption peak position and the number of carbonyl absorption peaks are extracted from the second matrix.

[0020] The preset standard threshold is compared with the number of carbonyl absorption peaks; when the number of carbonyl absorption peaks meets the preset first indicator, it is judged to meet the surface quality requirements of the injection molded part and the output is qualified; when the number of carbonyl absorption peaks does not meet the preset first indicator, the position of the carbonyl absorption peak is compared with the preset second indicator; when the position of the carbonyl absorption peak meets the preset second indicator, it is judged to meet the surface quality requirements of the injection molded part and the output is qualified; if the position of the carbonyl absorption peak does not meet the preset second indicator, it is judged to not meet the surface quality requirements of the injection molded part and the output is unqualified.

[0021] Furthermore, the method of performing Fourier transform on the first spectrum and the second spectrum respectively, and converting the time domain interference spectrum into frequency domain spectrum data to obtain the first spectrum data and the second spectrum data includes: A Blackman-Harris window function is applied to the time-domain interference spectrum data of the first spectrum and the second spectrum respectively; and a phase error function is calculated using a Mertz phase correction algorithm with the He-Ne laser interferogram as a reference.

[0022] The first and second spectrum images with the Blackman-Harris window function are respectively subjected to FFT calculations using the radix-2 Cooley-Tukey algorithm. The time domain interference spectrum is converted into frequency domain spectrum data using the spectrum conversion formula to obtain the first spectrum data and the second spectrum data. The spectrum conversion formula is calculated as follows: ; in, is the interference pattern intensity, is the optical path difference, is the spectrum data, is the wave number.

[0023] Furthermore, the method of performing smooth second-order derivative processing on the fitted first spectral data and the second spectral data to separate overlapping peaks to obtain the first characteristic spectral data and the second characteristic spectral data includes: The fitted first spectrum data and the second spectrum data are calculated respectively by the first derivative algorithm. The first derivative algorithm includes a first derivative formula using a nine-point smoothing window and a quadratic polynomial. The calculation formula of the first derivative formula is: ; in, The spectral data is in wavenumber The second derivative at , represents the differential operation for calculating the derivative of spectral data, is the weight coefficient, For the wave number The first derivative spectrum data value at , is an integer variable.

[0024] A local extreme value search is performed on the first spectrum data and the second spectrum data calculated by the first derivative algorithm, and the peak position and absorption peak parameters are extracted to obtain the first characteristic spectrum data and the second characteristic spectrum data.

[0025] Furthermore, the method of decomposing and coupling the first characteristic spectrum data based on the first fused characteristic vector to obtain characteristic parameters of the carbonyl absorption peak includes: Extract attention from the first fusion feature vector and construct the attention weight matrix through the attention extraction formula; the attention extraction formula is: ; in, Indicates the The attention head The attention weight on the feature dimension, is the attention weight matrix, is the index of the attention head, is the index of the feature dimension, Is the loop variable.

[0026] A projection matrix is ​​constructed to project the fused eigenvector into the carbonyl characteristic subspace; the characteristic parameters of the carbonyl absorption peak are obtained by mapping the eigenvector to the peak parameters through the multi-layer perceptron in the carbonyl characteristic subspace.

[0027] Furthermore, the method of constructing a non-negative matrix decomposition model based on the second characteristic spectrum data; the non-negative matrix decomposition model is the product of a first matrix and a second matrix, wherein the first matrix is ​​the characteristic parameters of the carbonyl absorption peak and the second matrix is ​​the absorption peak parameters of the flame retardant and the carbonyl group includes: Extract the spectral data of the 1650-1750cm⁻¹ band from the second characteristic spectrum data and construct a spectral matrix ,in, is the sampling number, and the wavelength interval is 1cm⁻¹.

[0028] Normalize the spectral matrix.

[0029] Set the basis matrix to , the coefficient matrix is ; The first column Initialized to the characteristic parameters of the carbonyl absorption peak, the second column Initialize the absorption peak parameters of flame retardant and carbonyl.

[0030] A non-negative matrix decomposition model was constructed using the characteristic parameters of the carbonyl absorption peak and the absorption peak parameters of the flame retardant and carbonyl.

[0031] Based on the same inventive concept, the present invention also provides a surface quality detection system for injection-molded parts of a laptop shell. The system includes a spectral data acquisition unit, a data preprocessing and feature extraction unit, and an analysis and evaluation unit connected in sequence.

[0032] The spectral data acquisition unit includes an infrared spectrometer, an optical sampler and a motion control subsystem.

[0033] The data preprocessing and feature extraction unit includes a signal conditioning circuit, a preprocessing algorithm module and a feature extraction module.

[0034] The parsing and evaluation unit includes a deep learning reasoning module, a non-negative matrix decomposition module and a quality assessment module.

[0035] (3) Beneficial effects Compared with the prior art, the present invention has the following beneficial effects: Through the synergistic effect of a non-negative matrix factorization algorithm and a Transformer fusion model, the carbonyl peak in the 1650-1750 cm⁻¹ band was separated from the flame retardant interference peak, effectively resolving the spectral overlap caused by the addition of flame retardants. Using a seventh-order polynomial and smoothed second-order derivative, the measurement error of the carbonyl peak position was further reduced, further improving the accuracy of feature parameter extraction compared to traditional Fourier transform infrared spectroscopy. By comparing the number and position of carbonyl absorption peaks with standard thresholds, the team further eliminated the problem of misidentification of qualified products as substandard due to deviations in the determination of carbonyl content in injection-molded laptop cases. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without making any creative efforts.

[0037] Figure 1 This is a flowchart of a method for detecting surface quality of injection-molded parts of a laptop shell according to an embodiment of the present invention; Figure 2 This is a module block diagram of a surface quality inspection system for injection-molded parts of a laptop shell according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Throughout this specification, references to "one embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with the embodiment or example is included in at least one embodiment of the present invention. Therefore, appearances of the phrases "one embodiment," "an embodiment," "an example," or "an example" in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in one or more embodiments or examples in any suitable combinations and / or subcombinations. Furthermore, it will be understood by those of ordinary skill in the art that the figures provided herein are for illustrative purposes only and are not necessarily drawn to scale. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0040] Example 1: Figure 1 As shown, this embodiment provides a method for detecting the surface quality of an injection-molded part of a notebook housing, comprising: Using an infrared spectrometer, the standard part is irradiated with a preset band according to preset parameters to obtain a first spectrum graph, and the injection molded part is irradiated with a preset band according to preset parameters to obtain a second spectrum graph; the first spectrum graph and the second spectrum graph are Fourier transformed to obtain first spectrum data and second spectrum data respectively.

[0041] The first spectral data and the second spectral data are preprocessed with baseline drift correction and intensity normalization, respectively; the preprocessed first spectral data and the second spectral data are feature extracted by polynomial fitting and derivative spectral analysis to obtain first characteristic spectral data and second characteristic spectral data, respectively; the first characteristic spectral data include carbonyl absorption peak parameters, and the second characteristic spectral data include absorption peak parameters of the flame retardant and the carbonyl group, absorption peak positions of the flame retardant and the carbonyl group, and the number of absorption peaks of the flame retardant and the carbonyl group.

[0042] The first characteristic spectrum data and the second characteristic spectrum data are input into the Transformer fusion model based on the attention mechanism; the first characteristic spectrum data is decomposed and coupled analyzed to obtain the characteristic parameters of the carbonyl absorption peak; the flame retardant absorption peak parameters and the carbonyl absorption peak parameters in the second characteristic spectrum data are separated by bringing the characteristic parameters of the carbonyl absorption peak and the second characteristic spectrum data into the non-negative matrix decomposition model to obtain the carbonyl absorption peak position and the number of carbonyl absorption peaks; based on the comparison of the carbonyl absorption peak position and the number of carbonyl absorption peaks with the standard threshold, a qualified or unqualified result is output.

[0043] Furthermore, the method of irradiating the injection molded part with an infrared spectrometer at a preset wavelength and according to preset parameters to obtain a first spectrum, and irradiating the injection molded part with a preset wavelength and according to preset parameters to obtain a second spectrum; and obtaining first spectral data and second spectral data by Fourier transforming the first spectrum and the second spectrum includes: The first spectrum was obtained by irradiating the standard sample multiple times with an infrared spectrometer using the 1650-1750 cm⁻¹ band and a scanning resolution of 2-4 cm⁻¹.

[0044] The second spectrum is obtained by irradiating the sample to be tested multiple times with an infrared spectrometer using the 1650-1750 cm⁻¹ band and a scanning resolution of 2-4 cm⁻¹.

[0045] Fourier transform is performed on the first spectrum image and the second spectrum image respectively, and the time domain interference spectrum image is converted into frequency domain spectrum data to obtain first spectrum data and second spectrum data.

[0046] Specifically, the 1650-1750 cm⁻¹ wavelength range was selected as the detection range, precisely covering the characteristic absorption regions of carbonyl groups (1720 cm⁻¹±20 cm⁻¹) and flame retardants (1660 cm⁻¹±15 cm⁻¹), while avoiding interference from other groups. A scanning resolution of 2-4 cm⁻¹ achieved a half-width separation of the carbonyl and flame retardant peaks of ≥1.5, a 30-50% improvement over conventional resolution, effectively preventing feature confusion caused by peak overlap. By accumulating 24-32 scans, the signal-to-noise ratio of the carbonyl peak at 1720 cm⁻¹ was increased from 15:1 to over 25:1, and the noise fluctuation of weak signals (absorbance ≤0.1 AU) was reduced by 30-40%.

[0047] Furthermore, the method of performing baseline drift correction and intensity normalization preprocessing on the first spectral data and the second spectral data, respectively; extracting characteristic parameters from the preprocessed first spectral graph and the second spectral graph by polynomial fitting and derivative spectral analysis, respectively, to obtain the first characteristic spectral data and the second characteristic spectral data includes: The ALS algorithm is used to perform baseline correction on the first and second spectral data respectively, and the parameter λ is set to 10 5 , p=0.01; the spectral intensities of the first spectral data and the second spectral data are mapped to the [0,1] interval through Min-Max normalization.

[0048] For the preprocessed first spectral data and second spectral data, the 1650-1750 cm⁻¹ band is defined as the analysis range, wherein the 1700-1740 cm⁻¹ band is preset as the carbonyl absorption peak region, and the 1650-1680 cm⁻¹ band is preset as the flame retardant absorption peak region.

[0049] The spectral curves in the first and second spectral data were fitted using a seventh-order polynomial. The fitting interval covered 50 cm⁻¹ on both sides of the target peak. The coefficients were optimized using the least squares method to make the sum of squares of the fitting residuals less than or equal to 0.005 AU².

[0050] The fitted first spectrum data and the second spectrum data are respectively smoothed by second-order derivative processing to separate overlapping peaks to obtain first characteristic spectrum data and second characteristic spectrum data.

[0051] Specifically, the ALS algorithm (λ=10 5, p=0.01) to effectively eliminate baseline drift caused by factors such as sample stage material and ambient temperature fluctuations, thereby reducing baseline flatness errors at 1720 cm⁻¹. Min-Max normalization maps spectral intensities to the [0,1] range, eliminating absorbance deviations caused by sample thickness variations (±0.2 mm). The RSD for intensity repeatability across the same batch of samples after normalization was ≤1.5%, a 20-35% improvement compared to unprocessed data. The analysis range was defined as 1650-1750 cm⁻¹, with the carbonyl absorption peak at 1700-1740 cm⁻¹ and the flame retardant interference peak at 1650-1680 cm⁻¹ focused on to avoid interference from other wavelengths (such as CH bending vibrations around 1500 cm⁻¹). The spectral curve was fitted using a seventh-order polynomial, covering a fitting interval of 50 cm⁻¹ on either side of the target peak. Least-squares optimization was used to optimize the coefficients to a residual sum of squares less than or equal to 0.005 AU², achieving over 90% peak shape restoration. Compared to a traditional cubic polynomial, this method reduced the full width at half maximum (FWHM) fitting error for the carbonyl peak at 1720 cm⁻¹ from 3.2 cm⁻¹ to 1.1 cm⁻¹. In particular, when the flame retardant addition level was ≥15%, the fitting accuracy of the overlapping peaks was improved by 40-60%.

[0052] During the surface quality inspection of laptop shell injection molded parts, the ALS algorithm is used to correct the baseline drift of infrared spectral data. The specific implementation is as follows: The ALS algorithm suppresses baseline drift by constructing a penalty function. Its core idea is to convert spectral data into Decomposition into real signals and baseline ,Right now The baseline fitting is achieved by minimizing the objective function: ;in, is the penalty factor, which controls the degree of baseline smoothing; is the derivative of the baseline, constraining the smoothness of the baseline. Specific parameter settings: penalty factor , balance signal fidelity and baseline smoothness to avoid overfitting; weight parameter By iteratively updating the weight matrix, the interference of strong absorption peaks in the spectrum on the baseline is suppressed. The input data is the spectrum data of the injection molded parts of the notebook shell collected by the infrared spectrometer (wavelength range 1650-1750cm⁻¹), including the frequency domain spectrum data converted from the time domain interference spectrum. Initialization baseline: Use linear interpolation to generate the initial baseline , as the starting point of iterative optimization. For each wave number point , calculate the current baseline Residuals from spectral data: ,like ( As the threshold, take 0.05AU), it is determined as the signal peak area, and the weight is set ; otherwise set 1. Reduce the impact of strong peak areas on the baseline through the weight matrix. When the baseline difference between two adjacent iterations is When , stop the iteration and output the final baseline .

[0053] Furthermore, the method of inputting the first characteristic spectrum data and the second characteristic spectrum data into a Transformer fusion model based on the attention mechanism; and decomposing and coupling the first characteristic spectrum data to obtain characteristic parameters of the carbonyl absorption peak includes: The first feature spectrum data and the second feature spectrum data are input into a 12-layer Transformer encoder architecture, where each layer contains 8 attention heads and a feedforward neural network.

[0054] Linear projection is performed on the input first characteristic spectrum data and the second characteristic spectrum data and sinusoidal position encoding is added to establish temporal correlation.

[0055] The attention weight is calculated through the Query, Key, and Value matrices. The first feature spectrum data is used as the Query and the second feature spectrum data as the Key / Value to calculate the cross-sample feature correlation and output the first fused feature vector.

[0056] Based on the first fused eigenvector, the first characteristic spectrum data is decomposed and coupled analyzed to obtain characteristic parameters of the carbonyl absorption peak.

[0057] Specifically, the Transformer encoder employs a 12-layer architecture with eight attention heads per layer, using a parallel computation mechanism to simultaneously capture the correlation features between the carbonyl peak (1720 cm⁻¹) and the flame retardant peak (1660 cm⁻¹) at different wavenumbers. This architecture achieves 2-3 times higher efficiency than traditional CNNs in modeling the feature interactions in the 1650-1750 cm⁻¹ band under 20% flame retardant interference. Through sinusoidal positional encoding and linear projection (projection matrix dimensions 768×6), the six-dimensional features (carbonyl peak intensity, half-width at half-maximum, flame retardant peak intensity, half-width at half-maximum, peak offset, and oxidation index) are mapped to a 768-dimensional feature space, establishing temporal correlations between the features. The output 768-dimensional first fused feature vector is decomposed using a decoupling matrix pre-trained using principal component analysis. The covariance between the carbonyl and flame retardant components is ≤ 0.03, achieving feature decoupling. The actual position error of the carbonyl peak reconstructed based on this vector is ≤±1.2cm⁻¹, and the peak area calculation error is ≤4%, effectively solving the peak shape distortion problem caused by flame retardants.

[0058] Furthermore, the method of separating the flame retardant absorption peak parameters and the carbonyl absorption peak parameters in the second characteristic spectrum data by substituting the characteristic parameters of the carbonyl absorption peak and the second characteristic spectrum data into a non-negative matrix decomposition model to obtain the carbonyl absorption peak position and the number of carbonyl absorption peaks; and outputting a qualified or unqualified result based on the comparison of the number of carbonyl absorption peaks and the carbonyl absorption peak position with the standard threshold value includes: The characteristic parameters of the carbonyl absorption peak are standardized; a non-negative matrix decomposition model is constructed based on the second characteristic spectrum data; the non-negative matrix decomposition model is the product of the first matrix and the second matrix, where the first matrix is ​​the characteristic parameters of the carbonyl absorption peak, and the second matrix is ​​the absorption peak parameters of the flame retardant and the carbonyl.

[0059] The non-negative matrix factorization model is trained through an iterative optimization algorithm, and the element values ​​in the first matrix and the second matrix are adjusted until convergence; the carbonyl absorption peak position and the number of carbonyl absorption peaks are extracted from the second matrix.

[0060] The preset standard threshold is compared with the number of carbonyl absorption peaks; when the number of carbonyl absorption peaks meets the preset first indicator, it is judged to meet the surface quality requirements of the injection molded part and the output is qualified; when the number of carbonyl absorption peaks does not meet the preset first indicator, the position of the carbonyl absorption peak is compared with the preset second indicator; when the position of the carbonyl absorption peak meets the preset second indicator, it is judged to meet the surface quality requirements of the injection molded part and the output is qualified; if the position of the carbonyl absorption peak does not meet the preset second indicator, it is judged to not meet the surface quality requirements of the injection molded part and the output is unqualified.

[0061] Specifically, Z-score normalization is performed on the characteristic parameters of the carbonyl absorption peak (peak position, intensity, and half-width at half maximum) to eliminate dimensional differences and the influence of outliers, ensuring the input stability of the non-negative matrix factorization model. The constructed decomposition model reconstructs the spectrum by multiplying the first matrix with the second matrix, enforcing the non-negative physical meaning. The objective function with a regularization term ensures that the basis matrix orthogonality index (OI) ≤ 0.08 and the reconstruction error (RE) ≤ 3.5%, improving separation accuracy by 50% compared to the unconstrained NMF algorithm and effectively addressing the aliasing problem of overlapping peaks. Using a multiplicative update rule for iterative solution, convergence is achieved within 250 iterations (objective function change rate < 0.005%). The position of the carbonyl peak at 1720 cm⁻¹ is extracted with an error of ≤ ±1.5 cm⁻¹, which is three times faster than the traditional gradient descent method. The number of carbonyl peaks (threshold 5-10) is used as the primary judgment indicator. When the limit is exceeded, the peak position (1720±5cm⁻¹) is used as the secondary indicator, forming a "quantity-position" double verification mechanism to effectively reduce the false positive rate of unqualified products.

[0062] Furthermore, the method of performing Fourier transform on the first spectrum and the second spectrum respectively, and converting the time domain interference spectrum into frequency domain spectrum data to obtain the first spectrum data and the second spectrum data includes: A Blackman-Harris window function is applied to the time-domain interference spectrum data of the first spectrum and the second spectrum respectively; and a phase error function is calculated using a Mertz phase correction algorithm with the He-Ne laser interferogram as a reference.

[0063] The first and second spectrum images with the Blackman-Harris window function are respectively subjected to FFT calculations using the radix-2 Cooley-Tukey algorithm. The time domain interference spectrum is converted into frequency domain spectrum data using the spectrum conversion formula to obtain the first spectrum data and the second spectrum data. The spectrum conversion formula is calculated as follows: ; in, is the interference pattern intensity, is the optical path difference, is the spectrum data, is the wave number.

[0064] Specifically, the radix-2 Cooley-Tukey algorithm combined with a Blackman-Harris window function performs a Fourier transform to convert the time-domain interferogram into a frequency-domain spectrum. Sidelobe suppression improves from -25dB to -57dB, thus preventing peak distortion caused by spectral leakage. Combined with He-Ne laser phase correction, the peak position error at 1720cm⁻¹ is controlled to ±1.0cm⁻¹, achieving a spectral resolution of 1.92cm⁻¹.

[0065] In addition, through the time domain interferogram Fourier transform and frequency domain spectrum The relationship is: ; When there is a phase error When , the actual spectrum can be expressed as: The Mertz algorithm uses the ideal interference pattern of He-Ne laser (wavelength 632.8nm) As a reference, calculate the phase correction function , to eliminate the phase distortion caused by factors such as instrument optical path offset and mirror tilt, the phase error function is defined as: ,in is the initial position of the moving mirror. The linear phase baseline is fitted by the least squares method to separate the nonlinear phase error component.

[0066] The radix-2 Cooley-Tukey algorithm is used to improve spectral resolution and computational efficiency. The Fourier transform of a point is decomposed into Level butterfly operations, each level contains Butterfly unit. For the time domain interferogram , its Fourier transform and frequency domain spectrum The relationship is: ; Through base-2 decomposition, the amount of operation is reduced from down to , which is suitable for real-time processing of infrared spectra. The time domain interference pattern of the laptop shell injection molded parts is sampled at equal intervals, and the optical path difference interval is , number of sampling points , covering the moving mirror range of ±1.31mm (corresponding to the maximum wave number resolution ). At the same time, the Blackman-Harris window function is applied to the time domain data, and its expression is: ,in 、 、 Used to suppress spectrum leakage. Rearrange according to the binary bit reversal rule, perform butterfly operation iteration, and then use the formula Frequency Index Convert to wave numbers , and finally through , and obtain the frequency domain spectrum .

[0067] Furthermore, the method of performing smooth second-order derivative processing on the fitted first spectral data and the second spectral data to separate overlapping peaks to obtain the first characteristic spectral data and the second characteristic spectral data includes: The fitted first spectrum data and the second spectrum data are calculated respectively by the first derivative algorithm. The first derivative algorithm includes a first derivative formula using a nine-point smoothing window and a quadratic polynomial. The calculation formula of the first derivative formula is: ; in, The spectral data is in wavenumber The second derivative at , represents the differential operation for calculating the derivative of spectral data, is the weight coefficient, For the wave number The first derivative spectrum data value at , is an integer variable.

[0068] A local extreme value search is performed on the first spectrum data and the second spectrum data calculated by the first derivative algorithm, and the peak position and absorption peak parameters are extracted to obtain the first characteristic spectrum data and the second characteristic spectrum data.

[0069] Specifically, a nine-point smoothing window and a quadratic polynomial first derivative algorithm are used to calculate the fitted spectral data, significantly enhancing the discernibility of the carbonyl peak at 1720 cm⁻¹ and the flame retardant peak at 1660 cm⁻¹. In the derivative spectrum, the carbonyl peak exhibits a distinct negative peak, while the flame retardant peak exhibits a positive peak. The two peak shapes differ significantly, and the separation can reach over 1.8, a 30-50% improvement over conventional derivative methods, effectively resolving the difficulty in distinguishing overlapping peaks. The nine-point smoothing window uses a weighted coefficient matrix to perform a weighted average of the spectral data, suppressing random noise (such as environmental interference) while retaining peak shape characteristics, achieving an optimal balance between noise reduction and accuracy. By locating the extreme points of the peaks through local extreme value search, parameters such as peak position, peak intensity, and half-height width can be accurately extracted.

[0070] The specific implementation of the first derivative algorithm for separating spectral characteristic peaks is as follows: The analysis range of 1650-1750 cm⁻¹ is selected, with the carbonyl absorption peak region of 1700-1740 cm⁻¹ and the flame retardant absorption peak region of 1650-1680 cm⁻¹ being eliminated, eliminating interference from other bands. A scanning resolution of 2-4 cm⁻¹ is used, ensuring a half-width separation of ≥1.5 between the carbonyl peak at 1720 cm⁻¹ and the flame retardant peak at 1660 cm⁻¹, ensuring peak discernibility during derivative calculations. The window weight matrix meets symmetry and normalization requirements, and random noise (e.g., absorbance fluctuations caused by environmental interference ≤0.01 AU) is suppressed by weighting the edge wavenumbers. For each wavenumber point coefficient, the spectral data of the four points before and after it (a total of nine points) are weighted and summed to form a smoothed spectral curve. The first-order derivative of the smoothed spectrum is calculated using the central difference method. The first-order derivative formula is: ;in (corresponding to the wavelength interval), so that the wave number resolution of the derivative spectrum is consistent with the original spectrum. Derivative the first-order derivative spectrum again to obtain the derivative spectrum At this time, the carbonyl peak shows an obvious negative peak (near 1720cm⁻¹), and the flame retardant peak shows a positive peak (near 1660cm⁻¹), and the separation degree between the two can reach more than 1.8.

[0071] Furthermore, the method of decomposing and coupling the first characteristic spectrum data based on the first fused characteristic vector to obtain characteristic parameters of the carbonyl absorption peak includes: Extract attention from the first fusion feature vector and construct the attention weight matrix through the attention extraction formula; the attention extraction formula is: ; in, Indicates the The attention head The attention weight on the feature dimension, is the attention weight matrix, is the index of the attention head, is the index of the feature dimension, Is the loop variable.

[0072] A projection matrix is ​​constructed to project the fused eigenvector into the carbonyl characteristic subspace; the characteristic parameters of the carbonyl absorption peak are obtained by mapping the eigenvector to the peak parameters through the multi-layer perceptron in the carbonyl characteristic subspace.

[0073] Specifically, an attention weight matrix is ​​extracted from the 768-dimensional first fused feature vector. A 64-dimensional feature weight distribution constructed through softmax normalization dynamically suppresses flame retardant interference features (for example, the weight of the feature dimension corresponding to the overlapping peak at 1680 cm⁻¹ is reduced to ≤0.12), while simultaneously enhancing carbonyl features (the weight of the dimension associated with 1720 cm⁻¹ is increased to ≥0.85). The fused feature vector is mapped to the carbonyl feature subspace using a pre-trained 128-dimensional projection matrix. Principal component analysis is used to constrain the covariance between the carbonyl component and the flame retardant component to ≤0.03, achieving feature decoupling. A three-layer MLP (128-dimensional input layer, 64-dimensional hidden layer, and 3-dimensional output layer) is used to achieve a nonlinear mapping from feature vectors to peak parameters. The peak position calculation error is ≤±1.2cm⁻¹ (compared to the traditional linear regression error of ±3.5cm⁻¹). The peak intensity correction accuracy reaches 0.001AU, a 20-30% improvement over polynomial fitting. The half-maximum width calculation deviation is ≤1.5cm⁻¹.

[0074] Furthermore, the method of constructing a non-negative matrix decomposition model based on the second characteristic spectrum data; the non-negative matrix decomposition model is the product of a first matrix and a second matrix, wherein the first matrix is ​​the characteristic parameters of the carbonyl absorption peak and the second matrix is ​​the absorption peak parameters of the flame retardant and the carbonyl group includes: Extract the spectral data of the 1650-1750cm⁻¹ band from the second characteristic spectrum data and construct a spectral matrix ,in, is the sampling number, and the wavelength interval is 1cm⁻¹.

[0075] Normalize the spectral matrix.

[0076] Set the basis matrix to , the coefficient matrix is ; The first column Initialized to the characteristic parameters of the carbonyl absorption peak, the second column Initialize the absorption peak parameters of flame retardant and carbonyl.

[0077] A non-negative matrix decomposition model was constructed using the characteristic parameters of the carbonyl absorption peak and the absorption peak parameters of the flame retardant and carbonyl.

[0078] Specifically, a 101×P spectral matrix was constructed by precisely extracting the 1650-1750 cm⁻¹ band from the second characteristic spectrum data. This focused on the characteristic absorption regions of carbonyl groups (1720 cm⁻¹) and flame retardants (1660 cm⁻¹), eliminating interference from other bands. A sampling density of 1 cm⁻¹ was used to ensure that the full width at half maximum (FWHM) of the 1720 cm⁻¹ peak had ≥8 sampling points. The spectral matrix was normalized, with intensity values ​​mapped to the [0, 1] range to eliminate absorbance deviations caused by sample thickness variations (±0.3 mm) and instrument fluctuations. The first column of the basis spectrum matrix W was initialized to the standard carbonyl peak characteristic parameters (generated by averaging 100 spectra of samples without flame retardant addition), and the second column was initialized to the standard basis spectrum with 20% flame retardant addition, ensuring that the decomposition model had clear physical meaning. This initialization method ensured a cosine similarity between the basis matrix and the actual spectrum of ≥0.95, preventing parameter misjudgment due to model divergence. A decomposition model with the formula V≈WH (where V is the spectral matrix, W is the basis matrix, and H is the coefficient matrix) was constructed, with non-negative constraints ensuring that the decomposition results conform to physical laws. The objective function with regularization (α=0.15, β=0.1) maintained the basis matrix orthogonality index (OI) ≤ 0.08 and the reconstruction error (RE) ≤ 3.5%. At a 20% flame retardant addition, the separation between the carbonyl and flame retardant peaks reached 1.8, a 50% improvement compared to the traditional unconstrained NMF algorithm, effectively resolving the issue of overlapping peaks.

[0079] In the surface quality inspection of injection-molded laptop casings, an iterative optimization algorithm combined with a non-negative matrix factorization (NMF) model was used to separate characteristic parameters to address the spectral overlap between flame retardant and carbonyl peaks. The specific implementation is as follows: The core of non-negative matrix decomposition is to transform the spectral matrix Decompose into basis matrix and coefficient matrix The product of , where: the first column Initialized to the characteristic parameters of the carbonyl absorption peak; the second column Initialized to the absorption peak parameters of flame retardant and carbonyl. The objective function of iterative optimization is: ;in, 、 is the regularization coefficient to suppress overfitting; 、 represents the Frobenius norm. At initialization, the first column Based on the spectrum mean of the standard sample without adding flame retardant, the characteristic parameters (peak position, intensity, half-height width) of the carbonyl peak at 1720 cm⁻¹ are locked; the second column Based on the standard base spectrum of 10-20% flame retardant addition, including the mixed characteristics of the 1660cm⁻¹ flame retardant peak and the 1720cm⁻¹ carbonyl peak, the coefficient matrix is ​​initialized by random values , and through the non-negativity constraint Ensure the physical meaning is reasonable. Update the basis matrix through iteration and coefficient matrix The element value of , each iteration follows the following update formula: Coefficient matrix update: ; Among them, the numerator term strengthens the correlation between the basis matrix and the spectral data, and the denominator term suppresses noise interference.

[0080] Basis matrix update: ; The basic matrix characteristics are dynamically adjusted through the mapping relationship between spectral data and coefficient matrix.

[0081] Convergence judgment conditions: The iteration stops when any of the following conditions is met: 1. The rate of change of the objective function ( is the objective function value); 2. The number of iterations reaches 150 times (empirical value, balancing computational efficiency and accuracy).

[0082] Based on the same inventive concept, the present invention also provides a surface quality detection system for injection-molded parts of a laptop shell. The system includes a spectral data acquisition unit, a data preprocessing and feature extraction unit, and an analysis and evaluation unit connected in sequence.

[0083] The spectral data acquisition unit includes an infrared spectrometer, an optical sampler and a motion control subsystem.

[0084] The data preprocessing and feature extraction unit includes a signal conditioning circuit, a preprocessing algorithm module and a feature extraction module.

[0085] The parsing and evaluation unit includes a deep learning reasoning module, a non-negative matrix decomposition module and a quality assessment module.

[0086] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0087] Finally, it should be noted that although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting the surface quality of injection-molded parts of a notebook shell, characterized in that: The method comprises: Step S1: using an infrared spectrometer, irradiating a standard part with a preset wavelength band and preset parameters to obtain a first spectrum, and irradiating an injection molded part with a preset wavelength band and preset parameters to obtain a second spectrum; performing Fourier transform on the first spectrum and the second spectrum to obtain first spectral data and second spectral data, respectively; Step S2, performing baseline drift correction and intensity normalization preprocessing on the first spectral data and the second spectral data, respectively; performing feature extraction on the preprocessed first spectral data and the second spectral data by polynomial fitting and derivative spectral analysis to obtain first characteristic spectral data and second characteristic spectral data, respectively; the first characteristic spectral data includes carbonyl absorption peak parameters, and the second characteristic spectral data includes absorption peak parameters of the flame retardant and the carbonyl group, absorption peak positions of the flame retardant and the carbonyl group, and the number of absorption peaks of the flame retardant and the carbonyl group; Step S3, inputting the first characteristic spectrum data and the second characteristic spectrum data into the Transformer fusion model based on the attention mechanism; decomposing and coupling the first characteristic spectrum data to obtain the characteristic parameters of the carbonyl absorption peak; separating the flame retardant absorption peak parameters and the carbonyl absorption peak parameters in the second characteristic spectrum data by bringing the characteristic parameters of the carbonyl absorption peak and the second characteristic spectrum data into the non-negative matrix decomposition model, and obtaining the carbonyl absorption peak position and the number of carbonyl absorption peaks; outputting a qualified or unqualified result based on the comparison of the number of carbonyl absorption peaks and the position of the carbonyl absorption peak with the standard threshold.

2. The surface quality inspection method of a laptop shell injection molded part according to claim 1, characterized in that: The method of irradiating the injection molded part with an infrared spectrometer at a preset wavelength and parameters to obtain a first spectrum, and irradiating the injection molded part with a preset wavelength and parameters to obtain a second spectrum; and obtaining first spectrum data and second spectrum data by Fourier transforming the first spectrum and the second spectrum includes: The first spectrum is obtained by irradiating the standard sample multiple times with an infrared spectrometer using a 1650-1750 cm⁻¹ band and a scanning resolution of 2-4 cm⁻¹; The second spectrum is obtained by irradiating the sample to be tested multiple times with an infrared spectrometer using a 1650-1750 cm⁻¹ band and a scanning resolution of 2-4 cm⁻¹; Fourier transform is performed on the first spectrum image and the second spectrum image respectively, and the time domain interference spectrum image is converted into frequency domain spectrum data to obtain first spectrum data and second spectrum data.

3. The surface quality inspection method of a laptop shell injection molded part according to claim 1, characterized in that: The method of performing baseline drift correction and intensity normalization preprocessing on the first spectral data and the second spectral data, respectively; and extracting characteristic parameters from the preprocessed first spectral graph and the second spectral graph by polynomial fitting and derivative spectral analysis to obtain the first characteristic spectral data and the second characteristic spectral data comprises: The ALS algorithm is used to perform baseline correction on the first and second spectral data respectively, and the parameter λ is set to 10 5 , p=0.01; the spectral intensities of the first spectral data and the second spectral data are mapped to the [0,1] interval through Min-Max normalization; For the pre-processed first spectral data and second spectral data, the 1650-1750 cm⁻¹ band is defined as the analysis range, wherein the 1700-1740 cm⁻¹ band is preset as the carbonyl absorption peak region, and the 1650-1680 cm⁻¹ band is preset as the flame retardant absorption peak region; The spectral curves in the first and second spectral data were fitted using a seventh-order polynomial. The fitting interval covered 50 cm⁻¹ on both sides of the target peak. The coefficients were optimized using the least squares method to make the sum of squares of the fitting residuals less than or equal to 0.005 AU². The fitted first spectrum data and the second spectrum data are respectively smoothed by second-order derivative processing to separate overlapping peaks to obtain first characteristic spectrum data and second characteristic spectrum data.

4. The surface quality inspection method of a laptop shell injection molded part according to claim 1, characterized in that: The method of inputting the first characteristic spectrum data and the second characteristic spectrum data into the attention mechanism-based Transformer fusion model; and performing decomposition and coupling analysis on the first characteristic spectrum data to obtain characteristic parameters of the carbonyl absorption peak includes: The first and second feature spectrum data are input into a 12-layer Transformer encoder architecture, where each layer contains 8 attention heads and a feedforward neural network. Performing linear projection on the input first characteristic spectrum data and the second characteristic spectrum data and adding sinusoidal position coding to establish temporal correlation; Calculate the attention weight through the query, key, and value matrix; use the first feature spectrum data as the query and the second feature spectrum data as the key / value to calculate the cross-sample feature correlation and output the first fusion feature vector; Based on the first fused eigenvector, the first characteristic spectrum data is decomposed and coupled analyzed to obtain characteristic parameters of the carbonyl absorption peak.

5. The method for detecting surface quality of injection-molded parts of a laptop shell according to claim 1, characterized in that: The method comprises: substituting characteristic parameters of the carbonyl absorption peak and the second characteristic spectrum data into a non-negative matrix decomposition model to separate the flame retardant absorption peak parameters and the carbonyl absorption peak parameters in the second characteristic spectrum data, obtaining the carbonyl absorption peak position and the number of carbonyl absorption peaks; and outputting a qualified or unqualified result based on a comparison between the number of carbonyl absorption peaks and the carbonyl absorption peak position and a standard threshold value. The characteristic parameters of the carbonyl absorption peak are standardized; a non-negative matrix decomposition model is constructed based on the second characteristic spectrum data; the non-negative matrix decomposition model is the product of a first matrix and a second matrix, wherein the first matrix is ​​the characteristic parameters of the carbonyl absorption peak and the second matrix is ​​the absorption peak parameters of the flame retardant and the carbonyl group; The non-negative matrix factorization model is trained by an iterative optimization algorithm, and the element values ​​in the first matrix and the second matrix are adjusted until convergence; the carbonyl absorption peak position and the number of carbonyl absorption peaks are extracted from the second matrix; The preset standard threshold is compared with the number of carbonyl absorption peaks; when the number of carbonyl absorption peaks meets the preset first indicator, it is judged to meet the surface quality requirements of the injection molded part and the output is qualified; when the number of carbonyl absorption peaks does not meet the preset first indicator, the position of the carbonyl absorption peak is compared with the preset second indicator; when the position of the carbonyl absorption peak meets the preset second indicator, it is judged to meet the surface quality requirements of the injection molded part and the output is qualified; if the position of the carbonyl absorption peak does not meet the preset second indicator, it is judged to not meet the surface quality requirements of the injection molded part and the output is unqualified.

6. The method for detecting surface quality of injection-molded parts of a laptop shell according to claim 2, characterized in that: The method of performing Fourier transform on the first spectrum graph and the second spectrum graph respectively to convert the time domain interference spectrum graph into frequency domain spectrum data to obtain the first spectrum data and the second spectrum data includes: Applying a Blackman-Harris window function to the time-domain interferometric spectrum data of the first spectrum and the second spectrum respectively; calculating a phase error function using a Mertz phase correction algorithm with the He-Ne laser interferogram as a reference; The first and second spectrum images with the Blackman-Harris window function are respectively subjected to FFT calculations using the radix-2 Cooley-Tukey algorithm. The time domain interference spectrum is converted into frequency domain spectrum data using the spectrum conversion formula to obtain the first spectrum data and the second spectrum data. The spectrum conversion formula is calculated as follows: ； in, is the interference pattern intensity, is the optical path difference, is the spectrum data, is the wave number.

7. The method for detecting surface quality of injection-molded parts of a laptop shell according to claim 3, characterized in that: The method of performing smooth second-order derivative processing on the fitted first spectral data and the second spectral data to separate overlapping peaks to obtain the first characteristic spectral data and the second characteristic spectral data includes: The fitted first spectrum data and the second spectrum data are calculated respectively by the first derivative algorithm. The first derivative algorithm includes a first derivative formula using a nine-point smoothing window and a quadratic polynomial. The calculation formula of the first derivative formula is: ; in, The spectral data is in wavenumber The second derivative at , represents the differential operation for calculating the derivative of spectral data, is the weight coefficient, For the wave number The first derivative spectrum data value at , is an integer variable; A local extreme value search is performed on the first spectrum data and the second spectrum data calculated by the first derivative algorithm, and the peak position and absorption peak parameters are extracted to obtain the first characteristic spectrum data and the second characteristic spectrum data.

8. The method for detecting surface quality of injection-molded parts of a laptop shell according to claim 4, characterized in that: The method of decomposing and coupling the first characteristic spectrum data based on the first fused characteristic vector to obtain characteristic parameters of the carbonyl absorption peak includes: Extract attention from the first fusion feature vector and construct the attention weight matrix through the attention extraction formula; the attention extraction formula is: ； in, Indicates the The attention head The attention weight on the feature dimension, is the attention weight matrix, is the index of the attention head, is the index of the feature dimension, is the loop variable; A projection matrix is ​​constructed to project the fused eigenvector into the carbonyl feature subspace; the characteristic parameters of the carbonyl absorption peak are obtained by mapping the eigenvector to the peak parameters through the multi-layer perceptron in the carbonyl feature subspace.

9. The method for detecting surface quality of injection-molded parts of a laptop shell according to claim 5, characterized in that: The method of constructing a non-negative matrix decomposition model based on the second characteristic spectrum data; the non-negative matrix decomposition model is the product of a first matrix and a second matrix, wherein the first matrix is ​​the characteristic parameters of the carbonyl absorption peak and the second matrix is ​​the absorption peak parameters of the flame retardant and the carbonyl group, includes: Extract the spectral data of the 1650-1750cm⁻¹ band from the second characteristic spectrum data and construct a spectral matrix ,in, is the number of sampling times, with a wavelength interval of 1cm⁻¹; Normalize the spectral matrix; Set the basis matrix to , the coefficient matrix is ; The first column Initialized to the characteristic parameters of the carbonyl absorption peak, the second column Initialized to the absorption peak parameters of flame retardant and carbonyl; A non-negative matrix decomposition model was constructed using the characteristic parameters of the carbonyl absorption peak and the absorption peak parameters of the flame retardant and carbonyl.

10. A surface quality inspection system for injection-molded parts of a laptop shell, used to perform the method according to any one of claims 1 to 9, characterized in that: The system comprises a spectral data acquisition unit, a data preprocessing and feature extraction unit and an analysis and evaluation unit connected in sequence; The spectral data acquisition unit includes an infrared spectrometer, an optical sampler and a motion control subsystem; The data preprocessing and feature extraction unit includes a signal conditioning circuit, a preprocessing algorithm module and a feature extraction module; The parsing and evaluation unit includes a deep learning reasoning module, a non-negative matrix decomposition module and a quality assessment module.

Images