Active and passive composite laser processing back wall protection method and system

Abstract

The invention discloses an active and passive composite laser processing back wall protection method and system. The active and passive composite laser processing back wall protection method comprises the steps that characteristic signals in the laser processing process are collected in real time, wherein the characteristic signals comprise sound pressure signals, light intensity signals and vibration signals of the processing position of the upper surface of a processing object in the laser processing process; generating a time sequence signal feature vector based on the feature signal, and inputting the time sequence signal feature vector into a trained laser processing state judgment model to obtain a current laser processing state; and according to the current laser machining state and prior machining time statistical data, the laser machining process is stopped in good time. By combining a passive protection material and an active control method, high-quality laser hole machining is achieved while the back wall face of the cavity is prevented from being damaged.

Description

Technical Field

[0001] The present invention belongs to the technical field of precision laser extreme manufacturing, and particularly relates to a method and system for laser processing back wall protection with active and passive compounding. Background Art

[0002] Laser hole processing technology has been widely applied to the processing technologies of many thin-walled cavity parts such as the drilling of film cooling holes on aero-engine blades, the drilling of nozzle holes of automotive fuel injectors, and the precision cutting of heart stents. Due to the cavity structure inside the part, when the laser penetrates the upper layer of the material, overprocessing is likely to occur, which will damage the back wall opposite to the cavity of the part. Microcracks and stress concentration are likely to occur in the damaged defect area, bringing potential fatigue fracture risks to the part. With the continuous progress of the design technology of cavity parts and special-shaped functional holes, the size of the cavity becomes smaller and the shape of the special-shaped hole becomes more complex, increasing the difficulty of back wall damage protection.

[0003] The protection methods for the back wall surface damage of cavity parts can be divided into two categories: active protection and passive protection. The active protection method monitors the characteristic signals in the laser processing process in real time, judges the penetration state of the laser, and stops the laser processing process in a timely manner at the penetration moment. The passive protection method fills the cavity with a sacrificial protection layer resistant to laser ablation to prevent the laser from continuing to act on the back wall surface of the cavity after penetrating the upper layer of the material. The passive protection method is often accompanied by difficulties in removal after processing and problems of residues in the cavity. And when only using the active protection method, due to the influence of factors such as the fluctuation of the acquisition system, the delay of the control system, and the change of the processing environment, random decision-making delays will occur, and it is difficult to achieve a 100% protection rate.

[0004] Chinese Patent with the application number CN202111017143.X discloses a method and system for combined protection of the back wall of laser processing based on layer difference. The method includes: first, processing the part material specimen and the protection material specimen respectively and collecting characteristic signals, and constructing a decision model of the characteristic signals and the current processing material. Based on the collection of characteristic signals and model training of a single material specimen, a high decision accuracy can be obtained in the same dataset. However, there are differences in the characteristic signals released by the laser when processing the specimen and when processing the part. In addition, when the laser is in the critical penetration state, it will act on both the part material and the protection material at the same time, making it difficult for the model trained by a single material specimen to handle new data in the actual processing process, and there is a problem of poor model generalization ability. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method for laser processing back wall protection with active and passive compounding, which combines passive protection materials with an active control method to achieve high-quality laser hole processing while preventing damage to the back wall surface of the cavity.

[0006] To achieve the aforementioned invention objectives, the present invention adopts the following solutions: One aspect of the present invention provides a method for protecting the back wall in laser processing with a combination of active and passive means, including: Collecting in real time the characteristic signals during the laser processing, where the characteristic signals include: the sound pressure signal, the light intensity signal, and the vibration signal at the processing position on the upper surface of the processing object during the laser processing; Generating a time-series signal feature vector based on the characteristic signals and inputting it into a trained laser processing state judgment model to obtain the current laser processing state; Stopping the laser processing process in a timely manner according to the current laser processing state and the prior statistical data of the processing time.

[0007] Optionally, the construction process of the laser processing state judgment model includes: Setting up a penetration signal acquisition system for the calibration part, and carrying out data acquisition according to the data set division target; the calibration part is a sample part that is exactly the same as the formal processing part in terms of material, thickness, and clamping method; Simultaneously collecting the processing position characteristic signals and the penetration light signals during the laser processing through the signal acquisition system to establish labels for the data set; Extracting large-scale time-series features from the data frames, and based on the feature selection algorithm, using the processing state as the output data and the feature high-dimensional vector as the input data to construct a training data set with both separability and versatility; Training the laser processing state judgment model from the training data set based on the machine learning method.

[0008] Optionally, the construction process of the training data set includes: Performing high-pass filtering on the collected sound signal data, determining the penetration moment of the laser according to the moment of significant increase in the light signal data, and dividing the sound signal data into a period before processing starts, a period before penetration, and a period after penetration; Recording the processing time of multiple processes and calculating the mean, standard deviation, and the density function of the normal distribution; Assigning the classification label 0 to the data vectors in the period before penetration and the classification label 1 to the data vectors in the period after penetration to complete the calibration between the sound signal data and the laser processing state; Performing autocorrelation frame division on all the data in the period before penetration and the period after penetration, extracting large-scale time-series features from a single frame, including: descriptive statistics, basis functions, stability, autocorrelation functions, information entropy theory, physical non-linearity, model fitting features; and selecting the most significant first number of features through the maximum correlation and minimum redundancy algorithm to form a multi-dimensional time-series feature vector; Construct a training data set with the multi-dimensional time-series feature vectors as input data and the classification labels corresponding to each vector as output data.

[0009] Optionally, the timely stopping of the laser processing process according to the current laser processing state and prior processing time statistical data specifically includes: Start processing according to the processing flow, perform real-time frame division, filtering, and extraction of time-series features on the acoustic signal data collected during the processing, and then form multi-dimensional time-series feature vectors. Input the multi-dimensional time-series feature vectors at the most recent moment into the trained laser processing state judgment model to generate the processing state judgment result at that moment; After receiving the penetrated signal from the laser processing state judgment model, the laser processing control system calculates the weight value corresponding to the current penetration moment according to the density function of the normal distribution of the processing time. When three consecutive penetrated signals appear and their cumulative weight values are greater than the first cumulative threshold, it is determined that the current processing position has been penetrated; After continuing to process for the first preset time, stop the laser processing flow at the current processing position.

[0010] Optionally, the method further includes: filling the cavity of the processed part with passive protection materials, specifically: For cavities with a height less than the first preset height, use a circulating liquid with a flow rate of 0 to 1000 ml / min as the passive protection material. The flowing liquid is one or more of deionized water, carbon black ink, thickened ink, and organic solutions of laser absorption dyes; For cavities with a height greater than or equal to the first preset height, use a mixture of solid particles with a particle diameter of 0.3 mm to 1 mm and an adhesive for filling and curing as the passive protection material. The material of the solid particles is one or more of artificial graphite particles, tungsten carbide particles, alumina particles, zirconia particles, silica particles, silicon nitride particles, and artificial diamond crushed materials.

[0011] Another aspect of the present invention provides a main-passive composite laser processing back wall protection system, including: A laser processing process data acquisition module for real-time acquisition of characteristic signals during the laser processing process. The characteristic signals include: acoustic pressure signals, light intensity signals, and vibration signals at the processing position on the upper surface of the processing object during the laser processing process; A laser processing state judgment module for generating time-series signal feature vectors based on the characteristic signals and inputting them into the trained laser processing state judgment model to obtain the current laser processing state; A laser processing control module for timely stopping the laser processing process according to the current laser processing state and prior processing time statistical data.

[0012] Optionally, the construction process of the laser processing state judgment model includes: Set up a penetration signal acquisition system for the calibration part, and conduct data acquisition according to the dataset division target; the calibration part is a sample part with the same material, thickness, and clamping method as the formal processing part; Through the signal acquisition system, synchronously collect the processing position feature signal and the penetration light signal during the laser processing process, and establish labels for the dataset; Extract large-scale time series features from the data frame. Based on the feature selection algorithm, use the processing state as the output data and the feature high-dimensional vector as the input data to construct a training dataset with both separability and versatility; Based on the machine learning method, train the laser processing state judgment model from the training dataset.

[0013] Optionally, the construction process of the training dataset includes: Perform high-pass filtering on the collected acoustic signal data, determine the penetration moment of the laser according to the significant improvement moment of the optical signal data, and divide the acoustic signal data into an unprocessed period, an unpenetrated period, and a penetrated period; Record the processing time of multiple processes, and calculate the mean, standard deviation, and probability density function of the normal distribution of the time; Assign the data vector in the unpenetrated period the classification label 0, and assign the data vector in the penetrated period the classification label 1 to complete the calibration between the acoustic signal data and the laser processing state; Perform autocorrelation framing on all data in the unpenetrated period and the penetrated period, and extract large-scale time series features from a single frame, including: descriptive statistics, basis functions, stability, autocorrelation functions, information entropy theory, physical nonlinearity, and model fitting features; and select the most significant first number of features through the maximum correlation and minimum redundancy algorithm to form a multi-dimensional time series feature vector; Use the multi-dimensional time series feature vector as the input data and the classification label corresponding to each vector as the output data to construct a training dataset.

[0014] Optionally, stopping the laser processing process in a timely manner according to the current laser processing state and prior processing time statistical data specifically includes: Start processing according to the processing flow, perform real-time framing, filtering, and extraction of time series features on the collected acoustic signal data during the processing, and then form a multi-dimensional time series feature vector. Input the multi-dimensional time series feature vector at the nearest moment into the trained laser processing state judgment model to generate the processing state judgment result at that moment; After receiving the penetrated signal from the laser processing state judgment model, the laser processing control system calculates the corresponding weight value at the current penetration moment according to the density function of the normal distribution of the processing time. When three consecutive penetrated signals appear and the sum of their weight values is greater than the first cumulative threshold, it is determined that the current processing position has been penetrated. After continuing the processing for the first preset time, the laser processing process at the current processing position is stopped.

[0015] Optionally, the system further includes: A passive protection material filling module, which is used to fill the cavity of the processed part with passive protection materials. Specifically: for cavities with a height less than the first preset height, a circulating liquid with a flow rate of 0 - 1000 ml / min is used as the passive protection material, and the flowing liquid is one or more of deionized water, carbon black ink, thickened ink, and organic solutions of laser absorption dyes; for cavities with a height greater than or equal to the first preset height, solid particles with a particle diameter of 0.3 mm - 1 mm are mixed with an adhesive and filled and cured as the passive protection material, and the material of the solid particles is one or more of artificial graphite particles, tungsten carbide particles, alumina particles, zirconia particles, silica particles, silicon nitride particles, and artificial diamond crushed materials.

[0016] Compared with the prior art, the present invention has at least the following advantages: (1) The active protection method for back wall damage based on machine learning provided by the present invention extracts a large number of high-contrast time series signal features from the characteristic signals of the laser processing area, and performs feature selection for the model performance. Finally, a laser processing state judgment model with relatively high classification accuracy and certain generalization ability is trained. (2) The timely stop method for laser drilling using prior processing time information provided by the present invention realizes reasonable constraints on the machine learning model by introducing known processing time statistics, ensuring that the error of premature stop of the laser processing process due to model misjudgment will not occur. (3) The main and passive composite laser processing back wall protection method provided by the present invention can effectively eliminate the damage caused by laser penetration processing to the back wall of the part cavity, and at the same time ensure the processing quality of the part. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1Schematic diagram of the damage to the back wall of the cavity generated when the laser penetrates the machined cavity part; Figure 2 Calibration machining schematic diagram for calibrating the laser machining state before formal machining in a specific embodiment of the present invention; Figure 3 Formal machining schematic diagram of applying the active and passive composite protection method in a specific embodiment of the present invention; Figure 4 Schematic diagram for dividing the machining state of the sound pressure waveform in a specific embodiment of the present invention; Figure 5 Implementation flowchart in a specific embodiment of the present invention; Figure 6 Schematic diagram of the protection effect of the back wall of the part cavity after laser machining obtained in a specific embodiment of the present invention.

[0019] List of components and reference numerals: Specific embodiments

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Examples of these preferred embodiments are illustrated in the accompanying drawings. The embodiments of the present invention shown in the drawings and described according to the drawings are merely exemplary, and the present invention is not limited to these embodiments.

[0021] An active and passive composite laser machining back wall protection method provided by the present invention, the method comprising: performing laser machining on a machining object containing a passive protection material, collecting characteristic signals during the machining process, establishing a machining state judgment model by means of time series analysis method and statistical analysis of machining time, distinguishing the current laser machining state through the judgment model, and stopping the machining process in a timely manner; the passive protection material being a flowing liquid or solid particles; the machining object being a double-layer or multi-layer part of a thin-walled cavity type; the machining device used for laser machining being a multi-axis precision laser machining device; the collection process referring to converting the characteristic signals during the laser machining process into electrical signals and recording; the time series analysis method being a high-separability time series data feature extraction method based on statistical analysis technology; the machining time being the laser machining breakdown time without the passive protection material; the machining state being the material currently being laser machined; the machining state judgment model being a classification decision model with the time series signal feature vector as the input and the machining state as the output.

[0022] Optionally, for cavities with a height less than 2 mm, a circulating liquid with a flow rate of 0 - 1000 ml / min is used as the protective material. The liquid includes but is not limited to: deionized water, carbon black ink, thickened ink, and organic solutions of laser absorption dyes.

[0023] Optionally, the thickened ink is obtained by adding an anionic water-soluble thickener of type C-199 with a mass fraction of 0 - 1.2% to carbon black ink with a concentration greater than 1 g / ml.

[0024] Optionally, for cavities with a height greater than or equal to 2 mm, solid particles with a particle diameter of 0.3 mm - 1 mm are used as the passive protective material. The particle materials include but are not limited to: artificial graphite particles, flake graphite, tungsten carbide particles, alumina particles, zirconia particles, silica particles, silicon nitride particles, and crushed artificial diamond materials.

[0025] Optionally, to further increase the density of the solid filling material, the particle material is mixed with an adhesive for filling. After the protective material layer is fully cured, laser processing is carried out. The adhesives used include but are not limited to: polycarbonate, polyurethane, polyvinyl alcohol, polyacrylamide, ethyl cellulose, silica sol, and paraffin.

[0026] The described processing process depends on the degrees of freedom of the processing equipment and the design of the processing shape, including but not limited to: vertical processing, inclined processing, horizontal processing, rotary processing, and multi-axis linkage processing.

[0027] Optionally, the processing shape requires the use of a laser penetration processing technology during processing, including but not limited to: circular through-holes, elliptical through-holes, irregular through-holes, slit cutting, profiling cutting, and blanking cutting.

[0028] Optionally, the characteristic signal acquisition process generally uses a data acquisition card to transmit the characteristic signals of the processing process sensed by the sensor in the form of voltage signals for informationization.

[0029] Optionally, the data acquisition card has a sampling frequency of 150 kHz - 2 MHz, a sampling accuracy of 16 bit - 32 bit, and has a synchronous acquisition function.

[0030] Optionally, the characteristic signals in the processing process include the sound pressure signal, light intensity signal, and vibration signal at the upper surface processing position during the laser processing process.

[0031] Optionally, the sound pressure signal is sensed by a capacitive sound-electric transducer facing the processing position. The frequency response range of the transducer is 0.2 Hz - 20 kHz, the sensitivity is 4 mV / Pa - 100 mV / Pa, and preferably it is a free-field microphone; the micro voltage output by the transducer is amplified by a preamplifier to a higher voltage signal with a peak voltage of 20 V.

[0032] Optionally, the sensors that can be used for the light intensity signal include but are not limited to: photodiode sensors, CCD image sensors, CMOS image sensors, vacuum phototubes, and spectrometer probes.

[0033] Optionally, when using a photodiode sensor to build an optical signal sensing system, according to the type of processing laser, the cut-off frequency of the photodiode is 100 MHz to 1 GHz, the spectral response wavelength is 220 nm to 1100 nm, and the photosensitive surface area is 0.79 mm 2 ~8.04 mm 2 ; The photocurrent signal sensed by the photodiode is subjected to current / voltage conversion through a transimpedance amplifier with a gain of 0 to 100 kV / A, and a voltage signal of 0 to ±3.3 V is output.

[0034] Optionally, the vibration signal is sensed by a piezoelectric vibration acceleration sensor mounted on the surface of the processing object. The frequency response range of the sensor is 0.5 Hz to 15 kHz, the sensitivity is 1 pC / g to 30 pC / g, and the output charge signal is converted into a voltage signal through a charge amplifier.

[0035] Optionally, the establishment of the processing state judgment model includes at least the following steps: 1) Set up a penetration signal acquisition system for the calibration part; 2) Clamp the calibration part on the processing equipment, and set up a characteristic signal sensor system facing the laser processing position in the equipment processing area; 3) Process the part according to the formal processing flow. During the processing, synchronously collect the penetration signal in the cavity and the upper surface characteristic signal during the processing; 4) By comparing the mutation moments of the penetration signal and the characteristic signal, accurately record the breakdown processing time, statistically analyze multiple groups of processing time data, calculate descriptive statistics, and at the same time complete the marking of the processing state data set; 5) Perform autocorrelation frame division on the collected characteristic signals, extract various time series data features from the data within one frame, and form a high-dimensional feature vector; 6) Use the processing state as the output data and the high-dimensional feature vector as the input data to form a training data set; 7) Based on machine learning methods, train and establish a processing state judgment model from the training data set.

[0036] Optionally, the calibration part in step 1) is a sample part that is exactly the same as the formal processing part in terms of material, thickness, and clamping method. Determine whether to fill the cavity of the calibration part with passive protection material according to the division target.

[0037] Optionally, when the division target is the penetration state of the upper cavity material, the calibration part is not filled with passive protection material, but a penetration optical signal sensor is placed to sense the sudden change in light intensity in the cavity after the laser breaks through the upper material, and the processing time is recorded based on the moment of the sudden change.

[0038] Optionally, when the division target is the penetration state after filling the solid passive protection material, the calibration part is filled with passive protection material, and at the same time the back wall material is removed. A penetration optical signal sensor is placed directly below the calibration part at the processing position to sense the sudden change in light intensity after the laser breaks through the upper material and the protection material, and the processing time is recorded based on the moment of the sudden change.

[0039] Optionally, the penetration optical signal sensor has the same configuration as the characteristic optical signal sensor and is connected to the same set of data acquisition systems.

[0040] Optionally, the light receiving surface of the penetration optical signal sensor faces the cavity area of the part; for a narrow cavity with a height of less than 1 mm, the optical signal can be collected by leading out an optical fiber from the cavity and then connecting it to the sensor.

[0041] Optionally, to effectively estimate the processing time, the multiple sets of processing time data in step 4) should not be less than 5 sets, and the calculated descriptive statistics include but are not limited to: mean, variance, standard deviation, and their respective 95% confidence intervals.

[0042] Optionally, during calibration processing, continue processing for 10 s to 30 s after the laser penetrates the material, so as to fully record the characteristic signals during the entire laser processing process.

[0043] Optionally, before performing the time series feature extraction in step 5), perform chaos detection on the collected data to determine whether downsampling of the data is required; when chaotic factors are found in the data, perform downsampling to improve the signal-to-noise ratio of the data.

[0044] Optionally, for the sound pressure characteristic signal, before extracting the features, remove the background noise of the processing environment through high-pass filtering at 12 kHz to 15 kHz to highlight the signal changes during the laser processing process.

[0045] Optionally, the self-correlation frame division in step 5) is to use the self-correlation function to determine the period of the characteristic signal, so as to divide a single data frame.

[0046] Optionally, the sound signal time series features in step 5) include but are not limited to the following types: descriptive statistic features, basis function features, stability features, self-correlation function features, information entropy theory features, physical nonlinear features, model fitting features. Transform the original two-dimensional time-voltage data vector into a multi-dimensional time-feature data vector with high separability.

[0047] Optionally, to reduce the computing resource requirements of the model and improve the generalization of the model, in step 5), a feature selection algorithm is further used to select the top several most significant features. The feature selection algorithms used include, but are not limited to: ReliefF algorithm, maximum relevance and minimum redundancy algorithm, analysis of variance algorithm, chi-square test algorithm, kruskal-wallis test algorithm, sequential forward selection algorithm, sequential backward selection algorithm, random forest algorithm, genetic algorithm.

[0048] Optionally, in step 6), the data before the penetration time of the upper layer material is marked as unpenetrated, and the data after the penetration time of the passive protection material is marked as fully penetrated; the data between the two times is marked as partially penetrated.

[0049] Optionally, in the dataset formed in step 6), the data frames within the 95% confidence interval of the average processing time of the upper layer material obtained in step 4) are removed.

[0050] Optionally, the training dataset contains data samples of multiple complete processing processes.

[0051] Optionally, in step 7), the laser penetration state classification model is a classification algorithm based on machine learning. The machine learning algorithms that can be used include, but are not limited to: Naive Bayes algorithm, K-nearest neighbor algorithm, linear discriminant algorithm, support vector machine algorithm, decision tree algorithm, clustering algorithm, multi-layer neural network algorithm, one-dimensional convolutional neural network algorithm.

[0052] Optionally, when selecting a machine learning algorithm, in addition to examining the accuracy and precision of the algorithm, the computing time of the algorithm also needs to be considered. On the premise of ensuring the classification ability of the algorithm, an algorithm with the shortest possible computing time is selected.

[0053] Optionally, to ensure the processing quality of high-value parts, a penalty for misclassified samples is introduced during model training. For samples that are not penetrated but are judged to be penetrated, that is, false positive samples, a greater loss function weight is assigned to add a cost-sensitive feature to the algorithm.

[0054] The loss function is represented in the following form: ; Where: N is the total number of samples; y i is the i true label of the th sample; i is the predicted value of the model for the ωi is the weight of the i th sample; is the loss of a single sample, such as binary cross - entropy loss, mean square error, etc.

[0055] Optionally, the timely stop of the laser processing process is achieved by weighted accumulation of the judgment results of multiple groups of continuous input data, so as to generate a more robust judgment result.

[0056] Optionally, the weight value of the judgment result is taken from the normal distribution density function of the processing time.

[0057] Optionally, after the processing state judgment model and the timely stop method output that the upper material is penetrated, continue to process for a period of time to ensure the processing quality.

[0058] Optionally, the time of continuing to process after penetration will not cause the breakdown of the protective material, and the processing is immediately stopped when the processing state judgment model outputs a complete breakdown judgment.

[0059] The processing device used for the laser processing is a multi - axis precision laser processing equipment, and the types of processed materials are not restricted.

[0060] As shown in the Figure 1 accompanying figure, for the double - layer part 1 with a cavity, when the focused high - energy laser beam 3 acts on the processing position 2 for through - hole processing, if the laser fails to stop in time after penetrating the upper layer of the part or there is no sacrificial material to block it, damage 5 will be caused to the back wall of the part cavity 4. To prevent this situation, a main - passive composite laser processing back - wall protection method described in the present invention is proposed.

[0061] Example 1: S1. Install a high - speed photodiode sensor 702 with a maximum sensitive wavelength of 760 nm and a cut - off frequency of 1 GHz inside the part 1 to collect the optical signal 701 emitted into the cavity after the laser penetrates the upper layer of the part. Connect the sensor 702 to a transimpedance amplifier with a gain of 10 kV / A, and the output of the amplifier is sent to the optical signal acquisition system.

[0062] S2. As shown in the Figure 2 accompanying figure, regard the part installed with the optical signal acquisition system according to S1 as a calibrated part, clamp the calibrated part to the processing equipment, and set an acoustic - electric transducer 602 with a frequency response range of 10 Hz - 20 kHz in the direction facing the processing position 2. The transducer collects the acoustic signal 601 during the processing; the output of the transducer 602 is sent to the signal acquisition system. In this embodiment, the optical signal acquisition system and the acoustic signal acquisition system use different channels of the same set of data acquisition system.

[0063] This embodiment uses an analog signal data acquisition system with a sampling rate of 200 kHz, an analog signal input resolution of 24 bits. The system is connected to an industrial control computer on the Windows operating system platform through an RJ45 bus, and all subsequent data operations are completed based on the Windows platform.

[0064] S3. Process the parts according to the formal processing flow. During processing, synchronously collect the feature signal 601 and the penetration signal 701, and continue processing for 10 s after the laser penetrates the upper material.

[0065] S4. As shown in the appendix Figure 4 First, perform a 13.5 kHz high-pass filter on the collected acoustic signal data, then determine the laser penetration moment according to the significant improvement moment of the optical signal data, and then divide the acoustic signal data into an unprocessed period 901, an unpenetrated period 902, and a penetrated period 903.

[0066] Record the processing time of multiple processes, and calculate the mean, standard deviation, and probability density function of the normal distribution of the time used.

[0067] S5. Assign the data vectors in the unpenetrated period 902 the classification label 0, and assign the data vectors in the penetrated period 903 the classification label 1, thus completing the calibration between the acoustic signal data and the laser processing state.

[0068] S6. First, perform autocorrelation framing on all the data in the unpenetrated period 902 and the penetrated period 903, and extract a large number of time series features from a single frame, including: descriptive statistics, basis functions, stability, autocorrelation functions, information entropy theory, physical non-linearity, and model fitting features.

[0069] Then, select the most significant 8 features through the maximum correlation and minimum redundancy algorithm to form a time - time series feature multi-dimensional data vector.

[0070] Finally, use the multi-dimensional time series feature vector as the input data, and use the classification label corresponding to each vector as the output data to construct a training data set, and set 70% of the samples as the data set and the remaining 30% of the samples as the validation set.

[0071] S7. The laser processing state classification model in this embodiment is designed based on the support vector machine algorithm, and the training data set in S6 is used for model training.

[0072] The above S1 - S7 complete the marking and training of the laser processing state classification model before formal processing.

[0073] S8. Inject thickened ink 801 with a flow rate of 500 ml / min into the cavity of the formal part, and the mass fraction of the thickener is 1%.

[0074] S9. As shown in Figure 3 the attached figure, clamp the formal part to the processing equipment and set up the acoustic signal acquisition system same as that in S2.

[0075] S10. Start the processing according to the processing flow. For the acoustic signal data collected during the processing, perform frame segmentation, filtering, and extraction of time series features in real time in the manner of S4 and S6, and then form a time - time series feature multi - dimensional data vector. Input the data vector at the most recent moment into the trained laser processing state judgment model to generate the processing state judgment result at that moment.

[0076] S11. When the laser penetrates the upper layer material of the part, the flowing thickened ink 801 inside the cavity will act as the sacrificial layer 802 to prevent the laser beam 3 from damaging the inner wall of the cavity.

[0077] S12. After the laser processing control system receives the penetrated signal from the classification model, calculate the corresponding weight value at the current penetration moment according to the density function of the normal distribution of the processing time obtained in S4. When three consecutive penetrated signals appear and the sum of their weight values is greater than 3, the processing system determines that the current processing position 2 has been penetrated.

[0078] S13. Continue the laser processing for 2 s at the current processing position 2, then stop the laser processing flow at the current processing position 2 and transfer to the processing flow of the next processing position until the entire processing of the part is completed.

[0079] S14. Remove the formal part and rinse the cavity with a large amount of clean water to remove the residue of the thickened ink in the cavity.

[0080] The attached Figure 5 figure is the brief flowchart of this embodiment.

[0081] Embodiment 2: For parts with a cavity height greater than or equal to 2 mm, the technical solution implemented in this application can be changed according to the actual on - site conditions: fill solid particles into the cavity 4 as passive protection materials.

[0082] Based on Embodiment 1, Embodiment 2 is changed as follows: S`1. Remove the back wall material of the part 1, mix tungsten carbide particles with a diameter of 0.5 mm and polyvinyl alcohol binder and fill them into the cavity, and fully cure them as the protection material. Set up a high - speed photodiode sensor 702 with a maximum sensitive wavelength of 760 nm and a cut - off frequency of 1 GHz below the part 1 to collect the optical signal 701 emitted into the cavity after the laser penetrates the upper layer material and the protection material of the part. Connect the sensor 702 to a transimpedance amplifier with a gain of 10 kV / A, and the output of the amplifier is sent to the optical signal acquisition system.

[0083] At the same time, steps S1 to S4 of Example 1 were also performed to record the processing time of multiple groups of non-protective materials, and to calculate the mean of the time and the 95% confidence interval of the mean.

[0084] S`2. The part filled with protective material and installed with the penetrating light signal collection system at the bottom according to S`1 is regarded as a calibration part, and the calibration part is clamped to the processing equipment, and a characteristic light signal collection system 603 with the same configuration as the penetrating light signal collection system 702 described in S`1 is set in the direction facing the processing position 2. In this embodiment, the two optical signal collection systems use different channels of the same data collection system.

[0085] This embodiment uses an analog signal data acquisition system with a sampling rate of 200kHz and an analog signal input resolution of 24 bits. The system is connected to an industrial computer with a Windows operating system platform via an RJ45 bus, and all subsequent data operations are completed based on the Windows platform.

[0086] S`3. Process the parts according to the formal processing flow. During processing, the characteristic signal 601 and the penetration signal 701 are collected synchronously. After the laser penetrates the upper material and the protective material, the processing continues for 10 seconds.

[0087] S'4. Determine the laser penetration time according to the significant increase time of the penetration light signal data, and then divide the acoustic signal data into the period before processing begins, the period without penetration, and the period of complete penetration. At the same time, divide the period without penetration into the period without upper material penetration and the period of upper material penetration based on the mean processing time without protective materials in S'1, and remove the data frame where the 95% confidence interval of the mean processing time without protective materials is located. S`5. Assign classification label 0 to the data vectors in the period when the upper material has not penetrated, assign classification label 1 to the data vectors in the period when the upper material has penetrated, and assign classification label 2 to the data vectors in the period when the upper material has completely penetrated.

[0088] S`6. First, autocorrelation framing is performed on all data in the period when the upper material has not penetrated, the period when the upper material has penetrated, and the period when the upper material is completely penetrated, and a large number of time series features are extracted from a single frame, including: descriptive statistics, basis functions, stability, autocorrelation function, information entropy theory, physical nonlinearity, and model fitting characteristics.

[0089] Then, the eight most significant features are selected through the maximum correlation and minimum redundancy algorithms to form a time-series feature multidimensional data vector.

[0090] Finally, the multi-dimensional time series feature vector is used as input data and the classification label corresponding to each vector is used as output data to construct a training data set. 70% of the samples are set as the data set and the remaining 30% of the samples are set as the validation set.

[0091] S`7. The laser processing state classification model in this embodiment is designed based on the support vector machine algorithm, and the training data set in S`6 is used for model training.

[0092] The above S`1~S`7 complete the marking and training of the laser processing state classification model before formal processing.

[0093] S`8. Fill the cavity of the formal part with tungsten carbide particles with a particle diameter of 0.5 mm mixed with polyvinyl alcohol binder, and wait for sufficient curing.

[0094] S`9. As shown in the appendix Figure 3 Clamp the formal part to the processing equipment, and set the optical signal acquisition system the same as in S`2.

[0095] S`10. Start processing according to the processing flow. Extract the frame and time-series features of the optical signal data collected during the processing in the manner of S`4 and S`6 in real time, and then form a time-time series feature multi-dimensional data vector. Input the data vector at the nearest moment into the trained laser processing state judgment model to generate the processing state judgment result at that moment.

[0096] S`11. When the laser penetrates the upper layer material of the part, the solidified layer of tungsten carbide particles inside the cavity will act as the sacrificial layer 802 to prevent the laser beam 3 from damaging the inner wall of the cavity.

[0097] S`12. After receiving the penetrated signal from the classification model, the laser processing control system calculates the corresponding weight value at the current penetration moment according to the density function of the normal distribution of the processing time of the non-protective material obtained in S`1. When three consecutive upper layer material penetrated signals appear and their weight values accumulate to be greater than 3, the processing system determines that the upper layer material at the current processing position 2 has been penetrated.

[0098] S`13. After continuing to process for 3 s, stop the laser processing flow at the current processing position 2, and transfer to the processing flow of the next processing position until all processing of the part is completed.

[0099] S`14. Remove the formal part and put it into an ultrasonic cleaner. Heat the cleaning solvent to 40°C and ultrasonically clean for 60 min to remove the residue of tungsten carbide particles in the cavity.

[0100] Appendix Figure 6 is a physical photo of the cross-section of the part after laser through-hole processing using the method described in this application. The shape of the through-hole meets the design requirements, and there is no damage to the back wall surface of the cavity.

[0101] It should be noted that in the present invention, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or apparatus.

[0102] It should be understood that although this specification is described in terms of embodiments, not every embodiment contains only one independent technical solution. This narrative manner of the specification is merely for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions of each embodiment can also be appropriately combined to form other embodiments understandable to those skilled in the art.

Claims

1. An active-passive composite laser processing back wall protection method, characterized in that: include: Real-time collection of characteristic signals during laser processing, wherein the characteristic signals include: sound pressure signals, light intensity signals and vibration signals at processing positions on the surface of the processing object during laser processing; Generate a time series signal feature vector based on the feature signal, and input it into a trained laser processing state judgment model to obtain a current laser processing state; The laser processing process is stopped in a timely manner according to the current laser processing state and a priori processing time statistical data.

2. The active-passive composite laser processing back wall protection method according to claim 1 is characterized in that: The construction process of the laser processing state judgment model includes: A penetration signal acquisition system is set up for the calibration parts, and the targets are divided according to the data set to carry out data acquisition; the calibration parts are sample parts that are exactly the same as the formally processed parts in terms of material, thickness, and clamping method; Through the signal acquisition system, the processing position feature signal and the penetration light signal in the laser processing process are synchronously collected to establish labels for the data set; Extract large-scale time series features from the data frame, and based on the feature selection algorithm, use the processing status as output data and the feature high-dimensional vector as input data to construct a training data set that is both separable and versatile. Based on the machine learning method, the laser processing state judgment model is trained from the training data set.

3. The active-passive composite laser processing back wall protection method according to claim 2 is characterized in that: The process of constructing the training data set includes: Performing high-pass filtering on the collected acoustic signal data, determining the laser penetration time according to the significant improvement time of the optical signal data, and dividing the acoustic signal data into a period before processing begins, a period before penetration, and a period after penetration; Record the processing time of multiple processings, and calculate the mean, standard deviation, and density function of the normal distribution; Assigning classification label 0 to the data vector in the non-penetration period, and assigning classification label 1 to the data vector in the penetration period, thereby completing the calibration between the acoustic signal data and the laser processing state; Autocorrelation framing is performed on all data in the non-penetrated period and the penetrated period, and large-scale time series features are extracted from a single frame, including: descriptive statistics, basis functions, stability, autocorrelation functions, information entropy theory, physical nonlinearity, and model fitting features; and the most significant first number of features are selected through maximum correlation and minimum redundancy algorithms to form a multidimensional time series feature vector; A training data set is constructed using the multi-dimensional time series feature vector as input data and the classification label corresponding to each vector as output data.

4. The active-passive composite laser processing back wall protection method according to claim 3 is characterized in that: The method of stopping the laser processing process in a timely manner according to the current laser processing state and the prior processing time statistical data specifically includes: The processing is started according to the processing flow, and the acoustic signal data collected during the processing is framed, filtered and time-series features are extracted in real time to form a multi-dimensional time-series feature vector. The multi-dimensional time-series feature vector at the latest moment is input into the trained laser processing state judgment model to generate the processing state judgment result at that moment; After receiving the penetration signal from the laser processing state judgment model, the laser processing control system calculates the weight corresponding to the current penetration moment according to the density function of the normal distribution of the processing time. When three consecutive penetration signals appear and their accumulated weights are greater than the first accumulated threshold, it is determined that the current processing position has been penetrated. After continuing the processing for the first preset time, the laser processing flow of the current processing position is stopped.

5. The active-passive composite laser processing back wall protection method according to claim 1, characterized in that: The method further comprises: filling the cavity of the processed part with a passive protective material, specifically: For a cavity whose height is less than the first preset height, a circulating liquid with a flow rate of 0 to 1000 ml / min is used as a passive protective material, wherein the circulating liquid is one or more of deionized water, carbon black ink, thickened ink, and laser absorbing dye organic solution; For a cavity whose height is greater than or equal to a first preset height, solid particles with a particle diameter of 0.3 mm to 1 mm are mixed with an adhesive and filled and solidified as a passive protective material. The material of the solid particles is one or more of artificial graphite particles, tungsten carbide particles, aluminum oxide particles, zirconium oxide particles, silicon oxide particles, silicon nitride particles, and artificial diamond crushed materials.

6. An active-passive composite laser processing back wall protection system, characterized in that: include: The laser processing data acquisition module is used to collect the characteristic signals of the laser processing process in real time, and the characteristic signals include: the sound pressure signal, the light intensity signal and the vibration signal of the processing position on the surface of the processing object during the laser processing; A laser processing state judgment module is used to generate a time series signal feature vector based on the feature signal, and input it into a trained laser processing state judgment model to obtain a current laser processing state; The laser processing control module is used to stop the laser processing process in a timely manner according to the current laser processing state and a priori processing time statistics.

7. The active-passive composite laser processing back wall protection system according to claim 6, characterized in that: The construction process of the laser processing state judgment model includes: A penetration signal acquisition system is set up for the calibration parts, and the targets are divided according to the data set to carry out data acquisition; the calibration parts are sample parts that are exactly the same as the formally processed parts in terms of material, thickness, and clamping method; Through the signal acquisition system, the processing position feature signal and the penetration light signal in the laser processing process are synchronously collected to establish labels for the data set; Extract large-scale time series features from the data frame, and based on the feature selection algorithm, use the processing status as output data and the feature high-dimensional vector as input data to construct a training data set that is both separable and versatile. Based on the machine learning method, the laser processing state judgment model is trained from the training data set.

8. The active-passive composite laser processing back wall protection system according to claim 7, characterized in that: The process of constructing the training data set includes: Performing high-pass filtering on the collected acoustic signal data, determining the laser penetration time according to the significant improvement time of the optical signal data, and dividing the acoustic signal data into a period before processing begins, a period before penetration, and a period after penetration; Record the processing time of multiple processings, and calculate the mean, standard deviation, and density function of the normal distribution; Assigning classification label 0 to the data vector in the non-penetration period, and assigning classification label 1 to the data vector in the penetration period, thereby completing the calibration between the acoustic signal data and the laser processing state; Autocorrelation framing is performed on all data in the non-penetrated period and the penetrated period, and large-scale time series features are extracted from a single frame, including: descriptive statistics, basis functions, stability, autocorrelation functions, information entropy theory, physical nonlinearity, and model fitting features; and the most significant first number of features are selected through maximum correlation and minimum redundancy algorithms to form a multidimensional time series feature vector; A training data set is constructed using the multi-dimensional time series feature vector as input data and the classification label corresponding to each vector as output data.

9. The active-passive composite laser processing back wall protection system according to claim 8, characterized in that: The method of stopping the laser processing process in a timely manner according to the current laser processing state and the prior processing time statistical data specifically includes: The processing is started according to the processing flow, and the acoustic signal data collected during the processing is framed, filtered and time-series features are extracted in real time to form a multi-dimensional time-series feature vector. The multi-dimensional time-series feature vector at the latest moment is input into the trained laser processing state judgment model to generate the processing state judgment result at that moment; After receiving the penetration signal from the laser processing state judgment model, the laser processing control system calculates the weight corresponding to the current penetration moment according to the density function of the normal distribution of the processing time. When three consecutive penetration signals appear and their accumulated weights are greater than the first accumulated threshold, it is determined that the current processing position has been penetrated. After continuing the processing for the first preset time, the laser processing flow of the current processing position is stopped.

10. The active-passive composite laser processing back wall protection system according to claim 6, characterized in that: Also includes: The passive protective material filling module is used to fill the passive protective material in the cavity of the processed part, specifically: for a cavity with a height less than a first preset height, a circulating flowing liquid with a flow rate of 0~1000ml / min is used as the passive protective material, and the flowing liquid is one or more of deionized water, carbon black ink, thickened ink, and laser absorbing dye organic solution; for a cavity with a height greater than or equal to the first preset height, solid particles with a particle diameter of 0.3mm~1mm are mixed with an adhesive to be filled and solidified as the passive protective material, and the material of the solid particles is one or more of artificial graphite particles, tungsten carbide particles, aluminum oxide particles, zirconium oxide particles, silicon oxide particles, silicon nitride particles, and artificial diamond crushed materials.

Images