A method and system for real-time monitoring of internal defects in a laser cladding process

Abstract

The application discloses a kind of real-time monitoring method and system to internal defect of laser cladding process, including the sound signal in laser cladding process of acquisition amplification, and to sound signal wavelet packet transform and ensemble empirical mode decomposition, realize the noise reduction processing of sound signal;The sound signal collected is converted into the light signal that CCD type spectral analyzer can identify by sound wave sensor and light modulator;According to the light signal of each position obtained, construct data sample and analyze, establish the corresponding relationship of light signal and different kinds of defects and the position of different defects, and establish database by this;Model is established using deep belief neural network, and the model is trained using a large amount of experimental data in the above database, automatically extract signal features and identify laser cladding state, compared with the light signal in database, realize the recognition of internal defect type and defect position.The application can monitor the defect of laser cladding process in real time, so that the microstructure and performance of laser cladding product are more excellent.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of laser cladding equipment, in particular to a method and system for real-time monitoring of internal defects in a laser cladding process. BACKGROUND

[0002] In recent years, laser cladding technology has been widely used in industrial production and has attracted great attention. Laser cladding technology uses high-energy laser to melt alloy powder with excellent performance on the surface of the substrate, thereby forming a cladding layer with good performance on the surface of the substrate. The laser cladding technology can form metal parts in one step. After intelligent process control, the formed dense metal parts are nearly net-shaped and almost do not need subsequent processing, effectively realizing rapid and cladding 3D printing of metal parts, making the manufacturing of products faster, more personalized and diversified, and widely used in many fields to adapt to the trend of economic globalization. However, the quality stability of the laser cladding printed parts is poor. The laser molten pool is affected by multiple parameters, which can easily produce defects such as cracks and pores, seriously affecting the performance of the cladding layer surface. Therefore, it is of great significance to integrate intelligent technology into the laser cladding process and to perform online real-time intelligent diagnosis and control during the printing process to truly realize the rapid and cladding printing of dense metal parts.

[0003] The molten pool is a pool-shaped liquid area formed by melting metal droplets on the base material, and its measurement data contains important welding process information and is the main basis for online quality analysis. The existing detection technologies for laser cladding process include detection based on electrical signals, monitoring based on acoustic principles, monitoring based on spectral analysis, monitoring based on temperature field, and monitoring based on visual imaging. The current widely used visual molten pool method cannot detect defects without obvious characteristics from strong light, high reflection, and flying dust interference factors. Therefore, different monitoring methods must be combined for collaborative detection to achieve online quality detection.

[0004] Acoustic emission technology can analyze the nature of the acoustic emission source by determining the direction of the acoustic emission source, record the time and location of the acoustic emission, and thus make relevant risk assessment according to the rules of the acoustic emission source. The acoustic emission technology has unique potential advantages in non-destructive testing, has low environmental requirements, and can be monitored in very harsh conditions. SUMMARY

[0005] To solve the above problems, the present application provides a method and system for real-time monitoring of internal defects in a laser cladding process, which combines visual imaging monitoring and acoustic emission monitoring technology to more effectively achieve online monitoring.

[0006] The technical scheme adopted by the present application is as follows:

[0007] A method for real-time monitoring of internal defects in a laser cladding process, comprising the following steps:

[0008] Step S1, collecting and amplifying the acoustic signal in the laser cladding process, and performing wavelet packet transform and ensemble empirical mode decomposition on the acoustic signal to realize noise reduction processing of the acoustic signal;

[0009] Step S2, converting the collected acoustic signal into a light signal recognizable by a CCD spectrum analyzer through an acoustic sensor and a light modulator; constructing a data sample and analyzing according to the obtained light signal at each position, establishing the correspondence between the light signal and different types of defects and the position of different defects, and establishing a database accordingly;

[0010] Step S3, using a deep belief neural network to establish a model, training the model with a large amount of experimental data in the above database, automatically extracting signal features and identifying the laser cladding state, comparing and analyzing with the light signal in the database, and realizing the identification of the type and position of internal defects.

[0011] Further, the step S1 is specifically: collecting and amplifying the acoustic signal in the laser cladding process, using Matlab software to perform wavelet packet threshold denoising on the acoustic signal, and then performing EEMD denoising on the denoised signal to realize noise reduction processing of the acoustic signal.

[0012] Further, the step S2 is specifically: the processed acoustic signal is converted into a light signal recognizable by a CCD camera through a sound wave sensor and a light modulator in turn, analyzed by a light vector analyzer, and the correspondence between the light signal and different types of defects and the position of different defects is established, and a database is established accordingly.

[0013] Further, the step S3 is specifically: a deep belief neural network is used to establish a model, and a large amount of experimental data in the above database is used to train the model, and the steps of training the model include:

[0014] Step S31: unsupervised pre-training based on restricted Boltzmann machine, pre-training using CD-k algorithm, and iterative calculation of W, a, b values of three units of RBM1, RBM2 and RBM3, and W and b values of the last BP unit, which can directly use randomly initialized values to obtain the best weight value.

[0015] Step S32: supervised fine-tuning training, which needs to first use the forward propagation algorithm to get a certain output value from the input, and then use the back propagation algorithm to update the weight value and bias value of the network, and finally establish an ideal data model.

[0016] The application also provides a real-time monitoring system for internal defects in a laser cladding process, which comprises an acoustic emission monitoring system, an acoustic emission acquisition and processing module, a sensor module and an optical signal analysis module; the acoustic emission monitoring system adopts a three-channel acoustic emission monitoring system, which is arranged in a 120-degree triangle, each channel being a set of independently operating monitoring equipment, and each channel being positioned by a probe array positioning method; the acoustic emission acquisition and processing module is used for acquiring acoustic signals in the laser cladding process and performing noise reduction processing, converting the acoustic signals into optical signals recognizable by a CCD spectrum analyzer through the sensor module and performing analysis, analyzing the characteristics of the optical signals of each defect position and establishing a database; and the optical signal analysis module adopts a deep belief neural network to automatically extract signal characteristics and identify the laser cladding state, compares and analyzes the optical signals in the database, and realizes identification of the types and positions of internal defects.

[0017] Further, the acoustic emission acquisition and processing module comprises an acoustic signal acquisition unit and an acoustic signal processing unit, the acoustic signal acquisition unit is used for collecting acoustic signals generated in the laser cladding process and amplifying the acoustic signals, and the acoustic signal processing unit is used for performing wavelet packet transform and ensemble empirical mode decomposition on the acoustic signals to realize noise reduction processing of the acoustic signals.

[0018] Further, the sensor module comprises an acoustic wave sensor and a light modulator, the acoustic wave sensor is used for converting acoustic wave signals into electrical signals, and the light modulator is used for converting the electrical signals into optical signals recognizable by a CCD spectrum analyzer.

[0019] Further, the CCD spectrum analyzer adopts a CCD as a detector, is provided with a data processing and output unit, and three CCD spectrum analyzers need to be used in different directions in cooperation with the acoustic emission monitoring system, after which data fusion is performed on the data measured in different directions, and three-dimensional data of a target observation point are obtained.

[0020] Further, the optical signal analysis module comprises a database and a spectrum analysis unit; the database is used for storing and analyzing collected optical signal samples, and establishes a relationship between optical signals and different types of defects and different defect positions; and the spectrum analysis unit adopts a deep belief neural network to automatically extract signal characteristics and identify the cladding state, and realizes identification of the types and positions of defects.

[0021] Further, the system further comprises a man-machine interaction unit, the man-machine interaction unit comprises a display screen and a serial port, and is used for displaying the state of the instrument, internal defects and positions and measured spectral patterns.

[0022] Further, the system further comprises a power module, the power module comprises a 220V alternating current power supply and a 12V direct current power supply, wherein the 220V alternating current is used to support the host computer operation and control of the monitoring machine room, and the 12V power supply is used for power supply in the signal acquisition and processing process of the lower computer.

[0023] The present application has the following beneficial effects:

[0024] (1) The database of the present application simultaneously uses sound signals and spectral analysis data, compared with only using sound signals or spectral signals, the database will be more abundant.

[0025] (2) The synergistic development of acoustic emission and visual image technology improves the accuracy of detection and control data, the great abundance of the database makes the system more accurate in identifying the types and positions of internal defects, thereby the hardware of the laser cladding printing system is more and more perfect, the accuracy of process control is improved, and the quality stability of the printed parts is improved.

[0026] (3) The present application ingeniously combines the CCD camera with the spectral analyzer, which solves the problem that the CCD camera cannot detect defects without obvious characteristics from strong light, high reflection, flying dust and other interference factors.

[0027] (4) By using a three-channel acoustic emission monitoring system in different directions and a matching CCD spectral analyzer, the data of internal defects can be comprehensively obtained, and the accuracy is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The present application is a real-time online monitoring method for internal defects in the laser cladding process, which comprises the following steps:

[0029] Figure 2 The present application is a real-time online monitoring system for internal defects in the laser cladding process, which comprises the following steps:

[0030] Figure 3 The present application is a real-time online monitoring system for internal defects in the laser cladding process, which comprises the following steps: EMBODIMENT

[0031] The present application will be further described below with reference to the accompanying drawings.

[0032] As shown in the drawings, the present application is a real-time monitoring method for internal defects in the laser cladding process, which comprises the following steps: Figure 1

[0033] Step S1, collect and amplify the sound signals in the laser cladding process, and perform wavelet packet transform and ensemble empirical mode decomposition on the sound signals to realize noise reduction processing of the sound signals;

[0034] ​Step S2, the collected acoustic signals are converted into light signals recognizable by the CCD spectrum analyzer through the acoustic sensor and the light modulator; a data sample is constructed and analyzed according to the obtained light signals at each position, a corresponding relationship between the light signals and different types of defects and positions of the different defects is established, and a database is established accordingly;

[0035] Step S3, a model is established by using a deep belief neural network, the model is trained by using a large amount of experimental data in the database, signal features are automatically extracted, and a laser cladding state is recognized, comparison and analysis are performed on the light signals in the database, so that recognition of the types and positions of internal defects is realized.

[0036] In step S1, the acoustic signals in the laser cladding process are collected and amplified, wavelet packet threshold denoising is performed on the acoustic signals by using Matlab software, and EEMD denoising is performed on the denoised signals, so that denoising processing of the acoustic signals is realized.

[0037] In step S2, the processed acoustic signals are converted into light signals recognizable by the CCD camera through the acoustic sensor and the light modulator in sequence, analysis is performed by using the light vector analyzer, a corresponding relationship between the light signals and different types of defects and positions of the different defects is established, and a database is established accordingly.

[0038] In step S3, a model is established by using a deep belief neural network, the model is trained by using a large amount of experimental data in the database, and the training process of the model includes the following steps:

[0039] Step S31, unsupervised pre-training based on a restricted Boltzmann machine is performed, the CD-k algorithm is used for pre-training, the values of W, a, and b of three units of RBM1, RBM2, and RBM3 and the values of W and b of the last BP unit need to be iteratively calculated, the values of the random initialization are directly used, and the optimal weight value is obtained.

[0040] Step S32, supervised fine-tuning training is performed, in the training process, a certain output value is obtained from the input by using the forward propagation algorithm, the weight value and the bias value of the network are updated by using the back propagation algorithm, and finally an ideal data model is established.

[0041] As shown in FIG. Figure 2 A real-time monitoring system for internal defects in a laser cladding process includes an acoustic power supply module 1, an acoustic emission monitoring system, an acoustic emission collection and processing module 2, a sensor module 3, a light signal analysis module 4, and a human-computer interaction unit 5.

[0042] The power supply module 1 includes a 220V AC power supply and a 12V DC power supply, wherein the 220V AC power supply is used to support the operation and control of the upper computer in the monitoring room, and the 12V power supply is used for power supply in the signal collection and processing process of the lower computer.

[0043] As Figure 3 shown, the acoustic emission monitoring system in the application adopts a three-channel acoustic emission monitoring system, the three-channel acoustic emission monitoring system is arranged in a 120-degree triangle, each channel is a set of independently operating monitoring equipment, and each channel is positioned by a probe array positioning method to position the acoustic emission source.

[0044] The acoustic emission acquisition and processing module 2 is used for acquiring acoustic signals in the laser cladding process, and performing noise reduction processing, converting the acoustic signals into optical signals recognizable by the CCD spectrum analyzer through the sensor module 3 and performing analysis, analyzing the characteristics of the optical signals of each defect position, and establishing a database.

[0045] The acoustic emission acquisition and processing module 2 includes an acoustic signal acquisition unit and an acoustic signal processing unit, the acoustic signal acquisition unit is used for collecting acoustic signals generated in the laser cladding process and amplifying the acoustic signals, and the acoustic signal processing unit is used for wavelet packet transform and ensemble empirical mode decomposition on the acoustic signals, to realize noise reduction processing on the acoustic signals.

[0046] The sensor module 3 includes an acoustic wave sensor 31 and a light modulator 32, the acoustic wave sensor 31 is used to convert acoustic wave signals into electrical signals, and the light modulator 32 is used to convert electrical signals into optical signals recognizable by the CCD spectrum analyzer. The CCD spectrum analyzer uses CCD as a detector, and has a data processing and output unit, three CCD spectrum analyzers need to be used in different directions with the acoustic emission monitoring system, and then the data measured in different directions are fused, and three-dimensional data of the target observation point are obtained.

[0047] The optical signal analysis module 4 automatically extracts signal characteristics and identifies the laser cladding state by using a deep belief neural network, compares and analyzes the optical signals in the database, and realizes identification of the internal defect type and the defect position.

[0048] The optical signal analysis module 4 includes a database and a spectrum analysis unit, the database is used for storing and analyzing the collected optical signal samples, and establishing the relationship between the optical signals and different types of defects and different defect positions, and the spectrum analysis unit automatically extracts signal characteristics by using a deep belief neural network and identifies the cladding state, to realize identification of the defect type and the position.

[0049] The man-machine interaction unit 5 in the application includes a display screen and a serial port, and is used for displaying the state of the instrument, the internal defects and positions, and the measured spectrum pattern.

Claims

1. A method for real-time monitoring of internal defects in a laser cladding process, characterized in that, The method comprises the following steps: Step S1, collecting and amplifying the acoustic signals in the laser cladding process, using Matlab software to carry out wavelet packet threshold denoising on the acoustic signals, and then carrying out EEMD denoising on the denoised signals to realize the denoising processing of the acoustic signals; Step S2, converting the processed acoustic signals into light signals recognizable by a CCD camera through a sound wave sensor and a light modulator in sequence, analyzing through a light vector analyzer, establishing the corresponding relationship between the light signals and different types of defects and the positions of the different defects, and establishing a database based on the same; Step S3, establishing a model by using a deep belief neural network, training the model by using a large amount of experimental data in the database, automatically extracting signal features and identifying the laser cladding state, comparing and analyzing the light signals in the database, and realizing the identification of the types and positions of internal defects; the step of training the model comprises: Step S31: unsupervised pre-training based on a restricted Boltzmann machine, pre-training by using a CD-k algorithm, and iteratively calculating the values of W, a, and b of three units of RBM1, RBM2, and RBM3, and the values of W and b of the last BP unit, directly using the randomly initialized values, and obtaining the optimal weight value; Step S32: supervised fine-tuning training, in which a forward propagation algorithm is used to obtain a certain output value from the input, a back propagation algorithm is used to update the weight value and bias value of the network, and finally an ideal data model is established.

2. A real-time monitoring system for internal defects in a laser cladding process, characterized by: The acoustic emission monitoring system comprises a three-channel acoustic emission monitoring system, a sound emission collection and processing module, a sensor module, and a light signal analysis module; the three-channel acoustic emission monitoring system is arranged in a 120-degree triangular shape, and each channel is a set of independent monitoring equipment; the channels are arranged in a probe array positioning method to position the acoustic emission source; The sound emission collection and processing module comprises a sound signal collection unit and a sound signal processing unit; the sound signal collection unit is used to collect and amplify the sound signals generated in the laser cladding process; The sound signal processing unit is used to carry out wavelet packet transformation and ensemble empirical mode decomposition on the sound signals to realize the denoising processing of the sound signals; The sound emission collection and processing module is used to collect the sound signals in the laser cladding process, carry out denoising processing, convert the sound signals into light signals recognizable by a CCD spectrum analyzer through the sensor module, analyze the light signals, analyze the characteristics of the light signals at each defect position, and establish a database; The sensor module comprises a sound wave sensor and a light modulator; the sound wave sensor is used to convert the sound wave signals into electrical signals; and the light modulator is used to convert the electrical signals into light signals recognizable by a CCD spectrum analyzer; The CCD spectrum analyzer uses a CCD as a detector, is provided with a data processing and output unit, and is used in cooperation with the acoustic emission monitoring system at different positions; then, the data measured in different directions are fused, and three-dimensional data of the target observation point are obtained. The light signal analysis module automatically extracts signal features and identifies the laser cladding state by using a deep belief neural network, and compares and analyzes the light signals in a database to identify the types and positions of internal defects. The light signal analysis module includes a database and a spectrum analysis unit; the database is used to store and analyze the collected light signal samples, and establish the relationship between the light signals and different types of defects and different defect positions; the spectrum analysis unit automatically extracts signal features and identifies the cladding state by using a deep belief neural network, thereby realizing the identification of the defect type and position. The system further includes a human-computer interaction unit and a power module; the human-computer interaction unit includes a display screen and a serial port, and is used to display the state of the instrument, internal defects and positions, and measured spectrum patterns; the power module includes a 220V AC power supply and a 12V DC power supply, wherein the 220V AC power supply is used to support the operation and control of the upper computer of the monitoring machine room, and the 12V power supply is used to power the lower computer signal acquisition and processing process.

Images