Laser Near-Net Shaping Working Distance Monitoring Method and System Based on Data Fusion

Abstract

The present invention discloses a method and system for monitoring the working distance of laser near-net shaping based on data fusion. The method includes: acquiring acoustic emission signals and molten pool temperature signals, performing data preprocessing and vector fusion processing on them to obtain the fusion signal feature vector of laser near-net shaping; training the Gaussian naive Bayes regression model with the fusion signal feature vector of laser near-net shaping to obtain the trained Gaussian naive Bayes regression model; monitoring the working distance of laser near-net shaping based on the trained Gaussian naive Bayes regression model for acoustic emission signals and molten pool temperature signals to obtain the working distance value of laser near-net shaping. By using the present invention, it is possible to improve the accuracy of monitoring the working distance of laser near-net shaping by adopting a feature-level signal fusion method without being affected by occlusion. As a method and system for monitoring the working distance of laser near-net shaping based on data fusion, the present invention can be widely applied to the field of intelligent human-computer interaction technology.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent human-computer interaction, and particularly to a method and system for monitoring the working distance of laser near-net shaping based on data fusion. Background Art

[0002] Laser near-net shaping is a kind of additive manufacturing technology. Laser creates a molten pool in the area to be additively manufactured, and continuously conveys metal powder or filamentous materials to deposit layer by layer to form high-performance three-dimensional parts. During the manufacturing process, due to the accumulation of thermal stress, defects such as warping or collapse are likely to occur on the surface to be additively manufactured of the part. If the working path of the robotic arm remains unchanged, it will cause a deviation between the actual working distance and the theoretical distance between the optical head and the part to be additively manufactured, affecting the additive manufacturing quality and material utilization rate. In severe cases, interference between the working path of the robotic arm and the part to be additively manufactured may even occur, resulting in damage to the equipment and the part. Therefore, during the laser near-net shaping process, real-time monitoring of the working distance between the optical head and the part to be additively manufactured and timely adjustment of the movement path of the robotic arm are of great significance for improving the product quality and production safety of laser near-net shaping.

[0003] The combination of multiple existing off-axis digital cameras can accurately measure the working distance between the optical head and the part to be additively manufactured, but this measurement method has certain requirements for the lighting conditions and is easily affected by occlusion. In summary, the technical problems existing in the related technologies need to be improved. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a method and system for monitoring the working distance of laser near-net shaping based on data fusion, which can improve the accuracy of monitoring the working distance of laser near-net shaping by adopting a feature-level signal fusion method without being affected by occlusion.

[0005] The first technical solution adopted by the present invention is: A method for monitoring the working distance of laser near-net shaping based on data fusion, comprising the following steps:

[0006] Obtain the acoustic emission signal during the laser near-net shaping process and the molten pool temperature signal during the laser near-net shaping process;

[0007] Perform data preprocessing and vector fusion processing on the acoustic emission signal and the molten pool temperature signal to obtain a fusion signal feature vector of laser near-net shaping;

[0008] Train the Gaussian Naive Bayes regression model with the fusion signal feature vector of laser near-net shaping to obtain a trained Gaussian Naive Bayes regression model;

[0009] Based on the trained Gaussian Naive Bayes regression model, monitor the working distance of laser near-net shaping for acoustic emission signals and molten pool temperature signals, and obtain the working distance value of laser near-net shaping.

[0010] Further, the step of acquiring the acoustic emission signal and the molten pool temperature signal during the laser near-net shaping process specifically includes:

[0011] Place the acoustic emission probe at the bottom of the substrate to be additive manufactured, and apply ultrasonic coupling agent between the acoustic emission probe and the bottom of the substrate to be additive manufactured. According to the preset sampling rate, acquire the acoustic emission signal during the laser near-net shaping process;

[0012] Coaxially set the two-color pyrometer with the laser path of the molten pool, set the trigger start signal, and acquire the molten pool temperature signal during the laser near-net shaping process.

[0013] Further, the step of performing data preprocessing and vector fusion processing on the acoustic emission signal and the molten pool temperature signal to obtain the fusion signal feature vector of laser near-net shaping specifically includes:

[0014] Perform band-pass filtering on the acoustic emission signal to obtain the filtered acoustic emission signal;

[0015] Perform signal windowing on the filtered acoustic emission signal and the molten pool temperature signal respectively to obtain the segmented acoustic emission signal and the segmented molten pool temperature signal;

[0016] Calculate the time-domain features and frequency-domain features of the segmented acoustic emission signal and the segmented molten pool temperature signal respectively to obtain the acoustic signal feature vector and the molten pool temperature signal feature vector;

[0017] Perform vector fusion and normalization processing on the acoustic signal feature vector and the molten pool temperature signal feature vector to obtain the fusion signal feature vector of laser near-net shaping.

[0018] Further, the expression for performing signal windowing on the filtered acoustic emission signal and the molten pool temperature signal respectively is specifically as follows:

[0019]

[0020] In the above formula, hanning(·) represents the hanning window function, M represents the window length, and n represents the input signal data.

[0021] Further, the time-domain features and frequency-domain features specifically include mean, peak value, absolute mean, root mean square, absolute standard deviation, envelope standard deviation, kurtosis, skewness, average power, spectral kurtosis, spectral skewness, and spectral centroid.

[0022] Further, the step of training the Gaussian Naive Bayes regression model with the fusion signal feature vector obtained by laser near-net forming includes the following specific steps:

[0023] Assign data labels to the fusion signal feature vectors obtained by laser near-net forming to obtain fusion signal feature vectors with labels, where the data labels represent the working distance value between the optical head and the part to be additively manufactured;

[0024] Construct a Gaussian Naive Bayes regression model;

[0025] Input the fusion signal feature vectors with labels into the Gaussian Naive Bayes regression model for class prediction to obtain the predicted working distance value;

[0026] Determine the distance deviation value based on the predicted working distance value and the working distance value between the optical head and the part to be additively manufactured;

[0027] Based on the distance deviation value, evaluate the model performance through the root mean square error and the coefficient of determination, and output the trained Gaussian Naive Bayes regression model.

[0028] Further, the expression of the Gaussian Naive Bayes regression model is specifically as follows:

[0029]

[0030] In the above formula, y pred represents the regression prediction value, P(y i ∣X) represents the probability that the feature vector X belongs to the class y i , y i represents the class label, n represents the input signal data, and i represents the i-th class label.

[0031] The second technical solution adopted by the present invention is: a working distance monitoring system for laser near-net forming based on data fusion, including:

[0032] A first module for acquiring the acoustic emission signal during the laser near-net forming process and the molten pool temperature signal during the laser near-net forming process;

[0033] A second module for performing data preprocessing and vector fusion processing on the acoustic emission signal and the molten pool temperature signal to obtain the fusion signal feature vector of laser near-net forming;

[0034] A third module for training the Gaussian Naive Bayes regression model with the fusion signal feature vector of laser near-net forming to obtain the trained Gaussian Naive Bayes regression model;

[0035] The fourth module is used to monitor the working distance of laser near-net shaping based on the trained Gaussian Naive Bayes regression model for acoustic emission signals and molten pool temperature signals, and obtain the working distance value of laser near-net shaping.

[0036] The beneficial effects of the method and system of the present invention are as follows: By acquiring the acoustic emission signals and molten pool temperature signals in the process of laser near-net shaping, compared with the working distance monitoring based on paraxial vision, it is not affected by occlusion and can support the working distance detection in application scenarios such as inner wall repair. Furthermore, data preprocessing and vector fusion processing are performed on the acoustic emission signals and molten pool temperature signals to obtain the fusion signal feature vector of laser near-net shaping. The Gaussian Naive Bayes regression model is trained with the fusion signal feature vector of laser near-net shaping to obtain the trained Gaussian Naive Bayes regression model. Through the Gaussian Naive Bayes regression model, the working distance in the process of laser near-net shaping can be monitored in real time. Further, the feature-level signal fusion method is adopted, and the signal stability is better than that of single signals. Finally, based on the trained Gaussian Naive Bayes regression model, the working distance of laser near-net shaping is monitored for the acoustic emission signals and molten pool temperature signals, and the working distance value of laser near-net shaping is obtained, thereby improving the accuracy of working distance monitoring of laser near-net shaping. Description of the Drawings

[0037] Figure 1 is the flowchart of the steps of the method for monitoring the working distance of laser near-net shaping based on data fusion of the present invention;

[0038] Figure 2 is the structural block diagram of the system for monitoring the working distance of laser near-net shaping based on data fusion of the present invention;

[0039] Figure 3 is the schematic diagram of monitoring the working distance of laser near-net shaping provided by a specific embodiment of the present invention;

[0040] Figure 4 is the schematic diagram of the device for acquiring acoustic emission signals and molten pool temperature signals provided by a specific embodiment of the present invention. Detailed Embodiments

[0041] The following further describes the present invention in detail with reference to the drawings and specific embodiments. For the step numbers in the following embodiments, they are only set for the convenience of elaboration and explanation, and no limitation is imposed on the order between steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0042] Refer to Figure 1 , the present invention provides a method for monitoring the working distance of laser near-net shaping based on data fusion, and the method includes the following steps:

[0043] S100. Obtain the acoustic emission signal and the molten pool temperature signal during the laser near-net shaping process;

[0044] Specifically, place the acoustic emission probe at the bottom of the substrate to be added, and apply ultrasonic coupling agent between the acoustic emission probe and the bottom of the substrate to be added. According to the preset sampling rate, obtain the acoustic emission signal during the laser near-net shaping process;

[0045] In this embodiment, as Figure 4 shown, use the acoustic emission equipment to collect the acoustic emission signals of several single-track laser near-net shaping experiments at different working distances, and record the actual working distances of the experiments. The working distance of the single-track experiment is designed to be between 12 mm and 24 mm. The acoustic emission probe is clamped at the bottom of the substrate to be added with a fixture with a pre-tightening force, and an appropriate amount of ultrasonic coupling agent is applied to the contact surface. The sampling rate is set to sample_rate_AE, where sample_rate_AE is greater than twice the highest frequency in the frequency response range of the acoustic emission equipment. According to the Nyquist sampling theorem, during the process of analog / digital signal conversion, when the sampling frequency sample_rate_AE is greater than twice the highest frequency fmax in the signal (sample_rate_AE >= 2fmax), the digital signal after sampling completely retains the information in the original signal. That is, the sampling rate of the data acquisition card needs to be set. According to the sampling theorem, the sampling rate of the data acquisition card needs to be greater than twice the maximum value of the frequency response range of the acoustic emission probe. In industrial applications, to ensure the reliability of the signal, 2.5 times is usually adopted. Specifically, use the acoustic emission equipment to measure the acoustic emission signal during the laser near-net shaping process, and the sampling rate is set to 1 Mps.

[0046] Coaxially set the two-color pyrometer with the laser path of the molten pool, set the trigger start signal, and obtain the molten pool temperature signal during the laser near-net shaping process.

[0047] In this embodiment, use the colorimetric pyrometer equipment to collect the molten pool temperature signal during the laser near-net shaping process. The two-color pyrometer and the aforementioned acoustic emission equipment use the same trigger signal to start the data acquisition, and the sampling rate is the same as that of the aforementioned acoustic emission equipment, both being 1 Mps. The acoustic emission sensor in contact connection with the part to be added and the coaxially installed two-color pyrometer have no requirements for the light conditions and are not affected by occlusion, and are suitable for industrial application scenarios with occlusion such as inner wall repair. The coaxial setting of the two-color pyrometer with the laser path of the molten pool represents the signal path of the pyrometer, that is, the path from the molten pool to the pyrometer, which coincides with the laser action path.

[0048] S200. Perform data preprocessing and vector fusion processing on the acoustic emission signal and the molten pool temperature signal to obtain the fusion signal feature vector of laser near-net shaping;

[0049] S210. Perform band-pass filtering on the acoustic emission signal to obtain the filtered acoustic emission signal;

[0050] In this embodiment, band-pass filtering is performed on the acoustic emission signal to improve the signal-to-noise ratio. The pass frequency range of the band-pass filtering is from P1 to P2, where P1 is the lowest frequency 35 kHz in the frequency response range of the acoustic emission device, and P2 is the highest frequency 400 kHz in the frequency response range of the acoustic emission device.

[0051] S220. Perform signal windowing on the filtered acoustic emission signal and the molten pool temperature signal respectively to obtain the segmented acoustic emission signal and the segmented molten pool temperature signal;

[0052] In this embodiment, the filtered acoustic emission signal is segmented into acoustic signal windows with consistent time intervals, and the molten pool temperature signal is segmented into molten pool temperature signal windows with consistent time intervals. Among them, the length of the windowing is M, the overlap rate is 0.5, and the window function is selected as the hanning window. The specific function is as follows:

[0053]

[0054] In the above formula, hanning(·) represents the hanning window function, M represents the length of the windowing, and n represents the input signal data.

[0055] In addition, it should be noted that the actual time corresponding to the length of the molten pool temperature signal window should be the same as the time corresponding to the length of an acoustic emission signal window.

[0056] More specifically, when a material or component deforms or cracks under stress, it releases strain energy in the form of elastic waves, a phenomenon known as acoustic emission (AE). Using AE equipment to monitor changes in elastic waves on the surface of the part being added can reveal the collision of powder with the surface and the fluctuations in the melt pool. A colorimetric pyrometer can accurately determine the temperature of a heated object based on the ratio of the radiation intensity at two wavelengths emitted by the object. Monitoring a coaxial melt pool using a colorimetric pyrometer can accurately reflect the maximum melt pool temperature. When the working distance between the optical head and the part being added changes, the powder flow distribution and melt pool energy density change accordingly. Using AE equipment and pyrometers to receive AE signals and melt pool temperature signals during the laser near-net-shape forming process allows for real-time monitoring of the working distance. Because the raw signals from AE equipment and pyrometers only present their respective physical information in the form of time series, they cannot directly identify the working distance. Analyzing the time and frequency domain characteristics of the windowed signals can reduce the amount of data to be processed while retaining the majority of the valid information. The Gaussian Naive Bayesian regression model can perform statistical regression on signals based on their characteristics, enabling monitoring of working distance during the laser near-net-shape process. Compared to working distance monitoring based on paraxial vision, acoustic emission equipment and pyrometers are unaffected by obstructions and can support working distance detection in applications such as inner wall repair. Furthermore, the fused signal is more reliable than a single signal.

[0057] S230, respectively calculating the time domain characteristics and the frequency domain characteristics of the segmented acoustic emission signal and the segmented molten pool temperature signal to obtain an acoustic signal feature vector and a molten pool temperature signal feature vector;

[0058] In this embodiment, a total of 12 time domain features and frequency domain features are calculated in each acoustic signal window to obtain an acoustic signal feature vector with a dimension of 1*12, and a total of 12 time domain features and frequency domain features are calculated in each molten pool temperature signal window to obtain a molten pool temperature signal feature vector with a dimension of 1*12.

[0059] 12 time domain and frequency domain features, including mean, peak, absolute mean, root mean square (RMS), absolute standard deviation, envelope standard deviation, kurtosis, skewness, average power, spectrum kurtosis, spectrum skewness and spectrum center of gravity. The calculation formulas and physical meanings of each feature are shown in Table 1, where x i is the i-th sample value in the data window, N is the length of the data window, is the absolute mean, e i =|h(x i )∣,h(x i ) is the signal envelope after Hilbert transform, is the mean of the envelope, s is the standard deviation, f k is the frequency, S(f k ) is the intensity of the spectrum, S is the mean of the spectrum, δs is the standard deviation of the spectrum.

[0060] Table 1 Calculation formula data table of each feature

[0061]

[0062]

[0063] S240 , performing vector fusion and normalization processing on the acoustic signal feature vector and the molten pool temperature signal feature vector to obtain a fusion signal feature vector of laser near-net forming.

[0064] In this embodiment, the acoustic signal feature vector corresponding to the same time period is merged with the molten pool temperature signal feature vector in the feature dimension direction to obtain the feature vector of the fusion signal, and normalization is performed for each type of feature in the feature matrix composed of the fusion feature vector to obtain the normalized fusion signal feature vector.

[0065] The normalized expression is as follows:

[0066]

[0067] In the above formula, max represents the maximum value of a feature in the dataset, min represents the minimum value of a feature in the dataset, x represents a feature value, and x' represents the normalized feature value.

[0068] S300, performing data training on a Gaussian naive Bayesian regression model using a fusion signal feature vector of the laser near-net-shape to obtain a trained Gaussian naive Bayesian regression model;

[0069] Specifically, a data label is assigned to the fusion signal feature vector of laser near-net forming to obtain a labeled fusion signal feature vector, wherein the data label represents the working distance value between the optical head and the part to be added; a Gaussian naive Bayes regression model is constructed; the labeled fusion signal feature vector is input into the Gaussian naive Bayes regression model for category prediction to obtain a predicted working distance value; a distance deviation value is determined based on the predicted working distance value and the working distance value between the optical head and the part to be added; based on the distance deviation value, the model performance is evaluated through the root mean square error and the determination coefficient, and the trained Gaussian naive Bayes regression model is output.

[0070] In this embodiment, it should be noted that the Gaussian Naive Bayesian regression model is an improved model based on the Gaussian Naive Bayesian classification model. Based on a given feature vector, the model calculates the probability that the feature vector belongs to each category and outputs the sum of the product of the probability of each category and the category label as the regression value. The formula is as follows:

[0071]

[0072] In the above formula, y pred represents the regression predicted value, P(y i ∣X) represents the probability that the feature vector X belongs to the category y i of, y i represents the category label, n represents the input signal data, and i represents the i-th category label.

[0073] The core idea of the Gaussian Naive Bayes regression model is to assume that the conditional probability of each feature follows a Gaussian distribution (normal distribution). In the regression scenario, the model calculates the probability that the feature vector belongs to each category, and takes the sum of the products of the probability of each category and the category label as the regression value output. It has the following advantages:

[0074] 1) High computational efficiency: Due to the conditional independence assumption, the computational complexity of the model is low, which is suitable for large-scale data sets.

[0075] 2) Suitable for small sample data: Gaussian Naive Bayes regression can still perform well under small samples because it uses prior probabilities and simple probability distribution assumptions.

[0076] 3) Good robustness: The model is not very sensitive to noise and missing values in the input data.

[0077] Furthermore, the model performance can be evaluated by the Root Mean Squared Error (RMSE) and the coefficient of determination (R-squared, R2): RMSE is used to measure the deviation between the sample predicted value and the true value. The smaller the RMSE value, the higher the prediction accuracy of the model; on the contrary, the larger the RMSE value, the worse the prediction accuracy. And the coefficient of determination R 2 is used to evaluate the goodness of fit of the regression model. The closer the R 2 value is to 1, the better the fitting degree of the model; the closer it is to 0, the worse the fitting degree. The specific calculation methods of the evaluation indicators are as follows:

[0078]

[0079] In the above formula, RMSE represents the root mean squared error function, R 2 represents the coefficient of determination, N represents the number of prediction samples, y i represents the true value of the i-th sample, y' i represents the predicted value of the i-th sample, and y represents the average value of all sample true values.

[0080] S400. Monitor the laser near-net shaping working distance based on the trained Gaussian Naive Bayes regression model for acoustic emission signals and molten pool temperature signals, and obtain the laser near-net shaping working distance value.

[0081] In summary, as Figure 3 shown, in the embodiments of the present invention, first, an acoustic emission device is used to collect acoustic emission signals of several laser near-net shaping single-pass experiments at different working distances, and the actual working distances of the experiments are recorded; the acoustic emission signals are subjected to band-pass filtering; the filtered acoustic emission signals are segmented into acoustic signal windows with consistent time intervals; 12 time-domain features and frequency-domain features within each acoustic signal window are calculated to obtain an acoustic signal feature vector with a dimension of 1×12; a colorimetric pyrometer device is used to collect the molten pool temperature signals during the laser near-net shaping process; the molten pool temperature signals are segmented into molten pool temperature signal windows with consistent time intervals; 12 time-domain features and frequency-domain features within each molten pool temperature signal window are calculated to obtain a molten pool temperature signal feature vector with a dimension of 1×12; the acoustic signal feature vector and the molten pool temperature signal feature vector corresponding to the same period of time are merged in the feature dimension direction to obtain a feature vector of the fusion signal, and normalization is performed for each type of feature in the feature matrix composed of the fusion feature vectors to obtain a normalized feature vector of the fusion signal; each fusion signal feature vector is labeled, and the label is the working distance value between the optical head and the part to be additive manufactured; a Gaussian naive Bayes regression model is established, and regression training is performed on the data set composed of the fusion feature vectors, and the trained model and parameters are saved; an acoustic emission device is used to obtain the acoustic emission signals during the laser near-net shaping process on-site, and a colorimetric pyrometer device is used to obtain the molten pool temperature signals during the laser near-net shaping process on-site; the trained Gaussian naive Bayes regression model and parameters are used to automatically identify the acoustic emission signals and molten pool temperature signals on-site, and the working distance label of the fusion signal on-site is obtained.

[0082] Referring to Figure 2 , the laser near-net shaping working distance monitoring system based on data fusion includes:

[0083] The first module 201 is configured to obtain the acoustic emission signals during the laser near-net shaping process and the molten pool temperature signals during the laser near-net shaping process;

[0084] The second module 202 is configured to perform data preprocessing and vector fusion processing on the acoustic emission signals and the molten pool temperature signals to obtain a fusion signal feature vector of the laser near-net shaping;

[0085] The third module 203 is configured to perform data training on the Gaussian naive Bayes regression model through the fusion signal feature vector of the laser near-net shaping to obtain a trained Gaussian naive Bayes regression model;

[0086] The fourth module 204 is configured to monitor the working distance of the laser near-net shaping based on the trained Gaussian naive Bayes regression model for the acoustic emission signals and the molten pool temperature signals to obtain the working distance value of the laser near-net shaping.

[0087] The content in the method embodiments described above is applicable to the system embodiments herein. The functions specifically implemented in the system embodiments are the same as those in the method embodiments described above, and the beneficial effects achieved are also the same as those in the method embodiments described above.

[0088] The above is a specific description of the preferred embodiments of the present invention. However, the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A laser near-net-shape working distance monitoring method based on data fusion, characterized in that: The following steps are involved: Place the acoustic emission probe on the bottom of the substrate to be added, apply ultrasonic coupling agent between the acoustic emission probe and the bottom of the substrate to be added, and obtain the acoustic emission signal during the laser near-net shaping process according to the preset sampling rate; The two-color pyrometer is coaxially set with the laser path of the molten pool, and a trigger start signal is set to obtain the molten pool temperature signal during the laser near-net forming process; Performing band-pass filtering on the acoustic emission signal to obtain a filtered acoustic emission signal; The filtered acoustic emission signal and the molten pool temperature signal are respectively subjected to signal window processing to obtain a segmented acoustic emission signal and a segmented molten pool temperature signal; The time domain characteristics and frequency domain characteristics of the segmented acoustic emission signal and the segmented molten pool temperature signal are calculated respectively to obtain the acoustic signal feature vector and the molten pool temperature signal feature vector; The acoustic signal feature vector and the molten pool temperature signal feature vector are fused and normalized to obtain the fused signal feature vector of laser near-net-shape forming. Assigning a data label to the fusion signal feature vector of the laser near-net shaping to obtain a fusion signal feature vector with a label, wherein the data label represents a working distance value between the optical head and the part to be added; Construct a Gaussian Naive Bayes regression model; The labeled fusion signal feature vector is input into the Gaussian Naive Bayes regression model for category prediction to obtain the predicted working distance value; Determine a distance deviation value based on the predicted working distance value and the working distance value between the optical head and the part to be added; Based on the distance deviation value, the model performance is evaluated by the root mean square error and the coefficient of determination, and the trained Gaussian naive Bayes regression model is output; The laser near-net-shape working distance is monitored based on the trained Gaussian naive Bayesian regression model using acoustic emission signals and molten pool temperature signals to obtain the laser near-net-shape working distance value.

2. The laser near-net-shape working distance monitoring method based on data fusion according to claim 1 is characterized in that: The specific expressions for performing signal window processing on the filtered acoustic emission signal and the molten pool temperature signal are as follows: ； In the above formula, represents the hanning window function, Indicates the length of the window, Indicates the input signal data.

3. The laser near-net shaping working distance monitoring method based on data fusion according to claim 2, characterized in that: The time domain features and frequency domain features specifically include mean, peak, absolute mean, root mean square, absolute standard deviation, envelope standard deviation, kurtosis, skewness, average power, spectrum kurtosis, spectrum skewness and spectrum center of gravity.

4. The laser near-net shaping working distance monitoring method based on data fusion according to claim 3 is characterized in that: The expression of the Gaussian Naive Bayes regression model is as follows: ； In the above formula, represents the regression prediction value, Represents the feature vector Belong to category The probability of represents the category label, Represents the input signal data, Indicates the Class category label.

5. Laser near-net-shape working distance monitoring system based on data fusion, characterized by: Includes the following modules: The first module is used to place the acoustic emission probe on the bottom of the substrate to be added, apply ultrasonic coupling agent between the acoustic emission probe and the bottom of the substrate to be added, and obtain the acoustic emission signal during the laser near-net shaping process according to a preset sampling rate; The two-color pyrometer is coaxially set with the laser path of the molten pool, and a trigger start signal is set to obtain the molten pool temperature signal during the laser near-net forming process; The second module is used to perform bandpass filtering on the acoustic emission signal to obtain a filtered acoustic emission signal; The filtered acoustic emission signal and the molten pool temperature signal are respectively subjected to signal window processing to obtain a segmented acoustic emission signal and a segmented molten pool temperature signal; The time domain characteristics and frequency domain characteristics of the segmented acoustic emission signal and the segmented molten pool temperature signal are calculated respectively to obtain the acoustic signal feature vector and the molten pool temperature signal feature vector; The acoustic signal feature vector and the molten pool temperature signal feature vector are fused and normalized to obtain the fused signal feature vector of laser near-net-shape forming. The third module is used to assign data labels to the fusion signal feature vector of the laser near-net shaping to obtain a fusion signal feature vector with a label, wherein the data label represents the working distance value between the optical head and the part to be added; Construct a Gaussian Naive Bayes regression model; The labeled fusion signal feature vector is input into the Gaussian Naive Bayes regression model for category prediction to obtain the predicted working distance value; Determine a distance deviation value based on the predicted working distance value and the working distance value between the optical head and the part to be added; Based on the distance deviation value, the model performance is evaluated by the root mean square error and the coefficient of determination, and the trained Gaussian naive Bayes regression model is output; The fourth module is used to monitor the laser near-net forming working distance based on the trained Gaussian naive Bayesian regression model for the acoustic emission signal and the molten pool temperature signal to obtain the laser near-net forming working distance value.

Images