Laser paint removal acoustic monitoring method and device based on neural network

Abstract

The invention provides a laser paint removal acoustic monitoring method and device, and relates to the field of laser cleaning.The method comprises the steps that in the laser paint removal process, sound signals generated by the action between laser and a paint layer and between laser and a substrate are collected; performing noise reduction processing on the sound signal; and extracting time-frequency domain characteristics of the denoised signal, inputting the time-frequency domain characteristics into the target neural network model for prediction, and judging whether paint is removed cleanly or the substrate is damaged. According to the method, the acoustic signal of laser paint removal is analyzed based on the neural network, the paint removal effect is judged, the method is a novel judgment method, use is easy and convenient, and the paint removal effect can be accurately judged in real time. The method has the characteristics of low cost, high accuracy and convenience.

Description

Technical Field

[0001] The present invention relates to the field of laser cleaning, and in particular to a method and device for acoustic monitoring of laser paint removal based on a neural network. Technical Background

[0002] As a new type of cleaning technology, laser cleaning technology has been successfully applied in many cleaning fields. Laser paint removal is an important application field of laser cleaning technology. Compared with traditional paint removal methods, it has the advantages of no mechanical contact, low operating cost, green and pollution-free, and can achieve online cleaning and local cleaning. This has made it widely studied and has shown great prospects in important industrial fields such as aviation, construction and automobiles.

[0003] To avoid damaging the substrate or inadequate cleaning, it is particularly important to monitor the laser paint removal process in real time. Currently, there are monitoring methods based on laser induced breakdown spectroscopy and images, but there are problems such as expensive detection equipment or poor anti-interference.

[0004] The present invention solves the above problem by using sound signals. The sound signals generated during the laser paint removal process contain information related to the paint removal effect and can be used for real-time monitoring. This method has the characteristics of low cost and high accuracy. Summary of the invention

[0005] The purpose of the present invention is to provide a method and device for acoustic monitoring of laser paint removal based on a neural network, which collects sound signals generated by the interaction between the laser and the paint layer and substrate, reduces noise on the collected signals, extracts time-frequency domain features, and obtains a target data set with clean paint removal, unclean paint removal, and damaged substrate as labels; obtains a neural network model of laser paint removal effect based on the target data set training model; and predicts the paint removal effect by inputting the real-time collected sound signals. The method has the advantages of high accuracy, low cost, and convenience.

[0006] In the first aspect, the embodiment of the present invention provides an online acoustic monitoring method for laser paint removal based on a neural network, the method comprising: collecting the sound signal generated by the interaction between the laser and the paint layer and the substrate during the laser paint removal process; performing noise reduction processing on the sound signal; extracting the time-frequency domain features of the noise-reduced signal and inputting them into the target neural network model for prediction, to determine whether the paint removal is clean or damages the substrate. The construction of the target neural network model is obtained by training the time-frequency domain features of the sound signal generated during the laser paint removal process.

[0007] According to the neural network-based laser paint removal acoustic monitoring method of claim 1, it is characterized in that the collected sound signal is generated during the laser paint removal process.

[0008] According to the neural network-based laser paint removal acoustic monitoring method of claim 1, it is characterized in that the collected sound signals are subjected to noise reduction processing, including: analyzing the frequency domain characteristics of the noise and the sound signal during the paint removal process, and performing noise reduction processing through a high-pass filter.

[0009] According to the neural network-based laser paint removal acoustic monitoring method of claim 1, it is characterized by simultaneously analyzing the time-frequency domain characteristics of the sound signal.

[0010] According to the neural network-based laser paint removal acoustic monitoring method of claim 1, the characteristic is that the time-frequency domain characteristics of the collected sound signals will change with the process of laser paint removal.

[0011] According to the neural network-based laser paint removal acoustic monitoring method of claim 1, the classification neural network model is trained based on the time-frequency characteristics of the sound signal to obtain the target neural network model, including: based on the time-frequency domain characteristics, the classification neural network model is trained to obtain an initial neural network model; wherein the classification neural network model adopts a classification model based on a convolutional neural network.

[0012] In the second aspect, an embodiment of the present invention provides an acoustic monitoring device for laser paint removal, the device comprising: a sound collection module and a prediction module; the collection module is used to: collect sound signals generated by the interaction between the laser and the paint layer and the substrate during the laser paint removal process; the monitoring module is used to: receive the collected sound signals; reduce noise on the sound signals; and predict the laser paint removal status.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] The equipment is simple and cheap, the processing method is fast and accurate, the device is convenient, and real-time online monitoring can be implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of acoustic monitoring of laser paint removal based on neural network according to the present invention;

[0016] Figure 2 It is a frequency domain characteristic curve diagram of the noise and laser paint removal sound signal of the present invention;

[0017] Figure 3 is a diagram of a laser paint removal acoustic monitoring device based on a neural network according to the present invention;

[0018] Figure 4 This is a diagram of the user interaction interface of the laser paint removal acoustic monitoring system of the present invention. Specific implementation plan

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0020] See also Figure 1 , Figure 1 The following is a flowchart of acoustic monitoring of laser paint removal based on neural network provided by an embodiment of the present invention. The embodiment of the present application is described in detail. The method includes: step 100, step 120 and step 140.

[0021] Step 100: Collecting sound signals during laser paint removal;

[0022] Step 120: De-noise the collected sound signal to obtain a time-frequency domain feature map of the sound signal after de-noising;

[0023] Step 140: Input the time-frequency domain feature map into the target neural network model for real-time prediction to determine the laser paint removal status.

[0024] In this embodiment, the laser uses a fiber pulse laser with a wavelength of 1064nm and an output power range of 0-100W. The laser emitted by the laser is transmitted through an optical fiber. The laser passes through a collimator to become a parallel beam, and then is focused by a focusing mirror with a focal length of 400mm. Finally, after being reflected by a single-axis scanning galvanometer, it is vertically irradiated on the painted base plate to generate a sound signal; the painted base plate uses an aluminum plate, and the paint is blue automatic spray paint, and the paint is evenly sprayed on the surface of the substrate; in this embodiment, the sound signal is collected using a microphone of model Apex220, and the monitoring range is 20Hz-20kHz. An audio interface acquisition card of model Rubix22 is used for time domain signal acquisition.

[0025] The specific steps are as follows: the laser emits a laser beam focused on the surface of the paint board, the microphone is placed about 10 cm away from the paint board, the sound signal generated during the paint removal process is collected, and all noise signals are collected separately. The frequency domain signal can be obtained through fast Fourier transform.

[0026] The frequency domain characteristic curve of the collected sound signal is as follows: Figure 2As shown in the figure, the noise signal is attenuated by about 90% in the frequency band of 20Hz to 2×103Hz compared with the maximum value near 2×103Hz. In addition, in the frequency band of 20Hz to 2×103Hz, the change trend of the sound generated by the laser paint removal process is the same as the change trend of the noise. In the frequency band of 2×103Hz to 20×103Hz, the noise is close to zero, and the sound generated by the laser paint removal process presents a peak value. This shows that the influence of noise is mainly concentrated in the frequency band below 2×103Hz.

[0027] A high-pass filter with a cutoff frequency set to 2 × 103 Hz was used to reduce the noise of the collected laser paint removal sound signal.

[0028] Perform time-frequency domain analysis on the denoised sound signal, calculate the Mel-frequency cepstral coefficients, and obtain the Mel-frequency cepstral coefficient spectrum.

[0029] In a real-time example, the method of constructing the target neural network in step 140 may include: step 121 and step 122.

[0030] Step 121: Change the experimental parameters to obtain the Mel-frequency cepstrum coefficient spectrum under the conditions of clean paint removal, unclean paint removal and damaged substrate.

[0031] Step 122: Gray-scale the Mel-frequency cepstral coefficient spectrum as input to train the classification neural network model to obtain the target neural network model.

[0032] Among them, the target neural network model is a classification model based on convolutional neural network, and the structure is as follows: the input is a single-channel grayscale image. There are three convolution layers, each layer uses a 23×23 convolution kernel, the first and second layers have 16 filters, and the third layer has 32 filters, with batch normalization and activation functions to enhance the stability of training and introduce nonlinear characteristics. There is a pooling layer after each convolution layer, which performs pooling through a 2x2 window to reduce the size of the feature map. The fully connected layer uses 256 neurons to extract global features. The output layer is defined according to the number of categories. The present invention is divided into three categories: clean paint removal, unclean paint removal, and substrate damage. The activation layer uses a softmax function to probabilize the output, and finally the cross entropy loss is calculated through the classification layer, and the output is compared with the target label to obtain the prediction result.

[0033] In addition, if Figure 4 As shown, an embodiment of the invention further provides a laser paint removal acoustic monitoring device based on a neural network, and the device includes: a sound collection module and a prediction module.

[0034] The above-mentioned sound collection module includes: a microphone of model Apex220, with a monitoring range of 20Hz-20kH; and an audio interface acquisition card of model Rubix22.

[0035] The above prediction module uses NVIDIA Jetson nano, into which the trained model can be deployed for acoustic monitoring. The prediction module may include: storage controller, main control chip processor, peripheral interface, and display unit.

[0036] Among them, the memory is random access memory (RAM), which is used to store model parameters and intermediate data during neural network inference to ensure operating efficiency.

[0037] The main control chip processors mentioned above include GPU and CPU. GPU provides high-performance computing capabilities to accelerate deep learning model reasoning, especially convolution operations for processing acoustic data features. CPU is responsible for system control, audio preprocessing, and peripheral interface management.

[0038] The above peripheral interface is used to connect a microphone.

[0039] The above display unit is used to connect to a display device, such as Figure 4 As shown, an interactive interface is provided for the user to start and shut down the laser paint removal monitoring system and display the status of the laser paint removal in real time.

[0040] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some communication interfaces, devices or units, which can be electrical, mechanical or other forms. The functional modules in the embodiments of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0041] It should be noted that if the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0042] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0043] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A laser paint removal acoustic monitoring method based on a neural network, characterized in that: The following steps are involved: During the laser paint removal process, the sound signals generated by the interaction between the laser and the paint layer and substrate are collected; Perform noise reduction processing on the collected sound signals to eliminate the influence of environmental noise; Extract the time-frequency domain features of the denoised sound signal; input the time-frequency domain features into the target neural network model for real-time prediction; The target neural network model is constructed by: The collected sound signal is framed and the short-time Fourier transform is calculated. The spectrum is converted to the Mel frequency scale, the logarithm of the energy of each frequency band is taken, and then a discrete cosine transform is performed to obtain the Mel frequency cepstral coefficients. The output Mel frequency cepstral coefficient feature matrix is ​​represented by a time-frequency spectrum, and then the obtained spectrum is grayed; the obtained spectrum is marked as clean paint removal, unclean and damaged substrate, corresponding to different paint removal states, and then used as input to train a classification neural network model to obtain the above-mentioned target neural network model.

2. The laser paint removal acoustic monitoring method based on neural network according to claim 1 is characterized in that The collected sound signals are generated during the laser paint removal process.

3. The laser paint removal acoustic monitoring method based on neural network according to claim 1 is characterized in that The collected sound signal is subjected to noise reduction processing, including: The frequency domain characteristics of the sound signal during noise and paint removal were analyzed, and noise reduction was performed through a high-pass filter.

4. The laser paint removal acoustic monitoring method based on neural network according to claim 1 is characterized in that Simultaneously analyze the time and frequency domain characteristics of the sound signal.

5. The laser paint removal acoustic monitoring method based on neural network according to claim 1 is characterized in that The time-frequency domain characteristics of the collected sound signals will change during the laser paint removal process.

6. According to the neural network-based laser paint removal acoustic monitoring method of claim 1, the training of a classification neural network model based on the time-frequency characteristics of the sound signal to obtain a target neural network model comprises: Based on the time-frequency domain features, training a classification neural network model to obtain an initial neural network model; Among them, the classification neural network model adopts a classification model based on convolutional neural network.

7. A laser paint removal acoustic monitoring device, characterized in that: The device comprises: a sound collection module and a prediction module; The acquisition module is used to: collect sound signals generated by the interaction between the laser and the paint layer and the substrate during the laser paint removal process; The monitoring module is used to: receive the collected sound signal; reduce the noise of the sound signal; predict the laser paint removal status; wherein the target neural network is constructed in a manner including: framing the collected sound signal, calculating the short-time Fourier transform, converting the spectrum to the Mel frequency scale, taking the logarithm of the energy of each frequency band, and then performing a discrete cosine transform to obtain the Mel frequency cepstral coefficients, the output Mel frequency cepstral coefficient feature matrix is ​​represented by a time-frequency spectrum, and then the obtained spectrum is grayed; the obtained spectrum is marked as clean paint removal and unclean paint removal for different paint removal states, and then used as input to train a classification neural network model to obtain the above-mentioned target neural network model.

Images