WO2021068983A1 - 一种基于智能声信息识别的焊后焊缝冲击质量判别方法及系统 - Google Patents

一种基于智能声信息识别的焊后焊缝冲击质量判别方法及系统 Download PDF

Info

Publication number
WO2021068983A1
WO2021068983A1 PCT/CN2020/124189 CN2020124189W WO2021068983A1 WO 2021068983 A1 WO2021068983 A1 WO 2021068983A1 CN 2020124189 W CN2020124189 W CN 2020124189W WO 2021068983 A1 WO2021068983 A1 WO 2021068983A1
Authority
WO
WIPO (PCT)
Prior art keywords
impact
weld
acoustic signal
quality
processing
Prior art date
Application number
PCT/CN2020/124189
Other languages
English (en)
French (fr)
Inventor
华亮
蒋凌
顾菊平
卢成
张堃
曹科才
商亮亮
张齐
王胜锋
葛雨暄
凌子茜
缪嘉伟
Original Assignee
南通大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南通大学 filed Critical 南通大学
Priority to US17/288,560 priority Critical patent/US11927563B2/en
Publication of WO2021068983A1 publication Critical patent/WO2021068983A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/20—Metals
    • G01N33/207—Welded or soldered joints; Solderability
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K31/00—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups
    • B23K31/006—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups relating to using of neural networks
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K31/00—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups
    • B23K31/12—Processes relevant to this subclass, specially adapted for particular articles or purposes, but not covered by only one of the preceding main groups relating to investigating the properties, e.g. the weldability, of materials
    • B23K31/125—Weld quality monitoring
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
    • B23K—SOLDERING OR UNSOLDERING; WELDING; CLADDING OR PLATING BY SOLDERING OR WELDING; CUTTING BY APPLYING HEAT LOCALLY, e.g. FLAME CUTTING; WORKING BY LASER BEAM
    • B23K37/00—Auxiliary devices or processes, not specially adapted to a procedure covered by only one of the preceding main groups
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/04—Analysing solids
    • G01N29/045—Analysing solids by imparting shocks to the workpiece and detecting the vibrations or the acoustic waves caused by the shocks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/34—Generating the ultrasonic, sonic or infrasonic waves, e.g. electronic circuits specially adapted therefor
    • G01N29/348—Generating the ultrasonic, sonic or infrasonic waves, e.g. electronic circuits specially adapted therefor with frequency characteristics, e.g. single frequency signals, chirp signals
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4481—Neural networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/46—Processing the detected response signal, e.g. electronic circuits specially adapted therefor by spectral analysis, e.g. Fourier analysis or wavelet analysis
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
    • H04R3/00—Circuits for transducers, loudspeakers or microphones
    • H04R3/04—Circuits for transducers, loudspeakers or microphones for correcting frequency response
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/023—Solids
    • G01N2291/0234—Metals, e.g. steel
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/028—Material parameters
    • G01N2291/0289—Internal structure, e.g. defects, grain size, texture
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/26—Scanned objects
    • G01N2291/267—Welds
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/26—Scanned objects
    • G01N2291/267—Welds
    • G01N2291/2675—Seam, butt welding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
    • H04R1/00—Details of transducers, loudspeakers or microphones
    • H04R1/08—Mouthpieces; Microphones; Attachments therefor

Definitions

  • the invention relates to the field of mechanical control, in particular to a method for judging the impact quality of a welded seam after welding based on intelligent acoustic information recognition.
  • Ultrasonic impact equipment uses high-power energy to push the impact head to impact the surface of metal objects at a frequency of about 20,000 times per second.
  • the high-frequency, high-efficiency, and high-focus energy causes large compression and plastic deformation of the metal surface; at the same time, the ultrasonic impact changes
  • the original stress field is generated and beneficial compressive stress is generated; the metal surface temperature rises rapidly and cools rapidly under high-energy impact, which changes the surface metal structure of the action area and strengthens the impact site.
  • the processing speed, pressure, angle, steel type, thickness and other factors during the operation of the ultrasonic impact method all determine the quality of the treatment effect of the ultrasonic impact method, but these factors are difficult to measure and quantify during the operation. Because the time-frequency domain characteristics of the ultrasonic shock sound signal during operation contain a lot of information, these sound signal characteristics are a comprehensive manifestation of external factors during the operation and determine the quality of the ultrasonic shock.
  • the quality of ultrasonic impact acting on the weld seam directly determines the degree of residual stress relief in the weld seam. Therefore, it is urgent to find a method for judging the impact quality of post-weld welds based on intelligent acoustic information recognition.
  • the purpose of the present invention is to provide a method and system for identifying the impact quality of a post-weld weld seam based on intelligent acoustic information identification with accurate identification and low monitoring cost.
  • a method for judging the impact quality of welds after welding based on intelligent acoustic information recognition which is characterized in that it includes the following steps:
  • a resistance strain gauge is used as a sensitive element for measurement, and an indentation is made by impact loading at the center of the strain rose.
  • the strain increase in the elastic area outside the indentation area is recorded by the strain gauge, so as to obtain the true elasticity corresponding to the residual stress. Strain, find the magnitude of the stress;
  • step S3 Establish a multi-weight neural network model, and train the multi-weight neural network model using the acoustic signal sample set marked in step S2) to obtain a multi-weight neural network that can be used to determine the impact quality of the post-weld weld The internet;
  • Step 1 Take the four features of each acoustic signal collected by the training sample collection module as a feature vector.
  • Step 2 Find the minimum value in the N ⁇ N-dimensional matrix A, and the position corresponding to the subscript is the sequence number of the two closest acoustic signal characteristic sample points, which are marked as P 11 and P 12 , and use them The first neuron ⁇ 1 constructed by corresponding two acoustic signal characteristic sample points;
  • Step 3 Delete the points covered by the first neuron ⁇ 1 in the acoustic signal feature sample point set ⁇ A 1 , A 2 ,..., A N ⁇ , and in the remaining acoustic signal feature sample point set, Calculate the distance of each point to point P 11 and point P 12 , find the two points with the shortest distance, record them as P 21 and P 22 , use acoustic signal characteristic sample points P 21 and P 22 to construct the second nerve of MDOFNN Yuan, denoted as ⁇ 2 ;
  • Step 4 follow Step 3, continue to process the remaining acoustic signal characteristic sample points, calculate P i1 P i2 , construct the i-th neuron, denoted as ⁇ i ;
  • the algorithm iteratively obtains the coverage of the two multi-weighted neuron coverage areas of "Qualified Processing Quality” and "Unqualified Processing Quality", and calculates the coverage of the test sample and two multi-weighted neuron networks representing the quality of post-weld weld stress processing.
  • the Euclidean distance between the regions and the Euclidean distance between the "processing quality qualified" multi-weighted neuron coverage area is the type that is the case where the post-weld weld stress processing quality in the test sample is qualified, and the "processing quality”
  • the type of "unqualified" multi-weight neuron coverage area with a closer European distance is the case where the quality of the post-weld weld stress treatment in the test sample is unqualified;
  • step S4) Obtain the characteristic value of the acoustic signal of the impact treatment of the weld after welding to be judged, and input the characteristic value into the multi-weight neural network trained in step S4), and output the judgment result of the impact treatment quality of the weld to be judged after the weld , That is, it is judged whether the impact treatment of the weld after welding to be judged is "qualified processing quality” or "processing quality unqualified".
  • Step S1) The specific processing process is: controlling the tip of the ultrasonic impact gun to perform impact processing on the weld after welding with different processing pressures, processing speeds, processing angles and impact frequencies to obtain acoustic signals during the impact processing;
  • Use Fourier transform to convert acoustic signals in the time domain into acoustic signals in the frequency domain, and use Butterworth filters to filter the acoustic signals in the frequency domain;
  • the acoustic signal After filtering the acoustic signal, the acoustic signal is divided into frames to extract its short-term characteristics, and short-time windowing technology is selected for frame processing.
  • the window used is a Hamming window with a window length of 1024 and an overlap of 50%.
  • the short-term zero-crossing rate, short-term average amplitude, short-term energy and short-term zero-energy ratio of the sound signal can be extracted from the time domain;
  • the short-term zero-crossing rate refers to the number of times the signal passes the zero value in a frame of signal; the calculation formula of the short-term zero-crossing rate Z n is as follows:
  • n represents the current sampling time point
  • N is the length of the Hamming window
  • x ⁇ (m) represents the signal after x(m) is windowed
  • x(m) is the amplitude of the acoustic signal at time m value
  • the impact energy of the sound signal is an ultrasonic weld gun impact significant difference between short-term energy E n is calculated as follows:
  • short-term energy is the square of an acoustic signal and in the signal, short-time average magnitude M n is used to calculate the absolute values of the acoustic signal and to measure the variation width; short-term energy calculation formula M n as follows:
  • the short-term zero-energy ratio is the ratio of the zero-crossing rate and the short-term energy within a frame of signal.
  • the short-term zero-energy ratio ZER n is calculated as follows:
  • a special post-weld seam impact quality discrimination system based on intelligent acoustic information recognition method for post-weld seam impact quality discrimination which is characterized by: including: a sound signal acquisition hardware platform for collecting post-weld seam impact processing
  • the signal processing and feature extraction module is used to filter and preprocess the acoustic signal and calculate the feature value; the discrimination module is used to input the feature value calculated by the signal processing and feature extraction module into the multi-weight neural network And output the quality judgment result.
  • the multi-weight neural network is a multi-weight neural network that can be used to judge the impact quality of the weld after welding after training;
  • the sound signal acquisition hardware platform includes: an ultrasonic impact gun, a mobile operating platform, a weldment to be processed, a free-field microphone, a sound vibration analyzer, and a PC; wherein the position of the ultrasonic impact gun is fixed and the weldment to be processed is fixed On the mobile operating platform, the mobile operating platform can form a relative movement with the ultrasonic impact gun 1 along the length direction of the weldment to be processed.
  • the ultrasonic impact gun is The weld toe is processed for the residual stress of the weld; the free-field microphone is placed in a circle with the tip of the ultrasonic impact gun as the center and a radius of 1.5m, which is used to collect the analog signal of the sound during the whole process and transmit it to the sound and Vibration analyzer:
  • the sound vibration analyzer converts the received analog signal of the sound into a digitized time-domain sound signal; then the sound vibration analyzer transmits the received sound information to the PC, and the PC is in the form of a file Stored; the signal processing and feature extraction module and the judgment module are both set in the PC, and then the signal processing and feature extraction module performs filtering and preprocessing on the acoustic signal and calculates the characteristic value; finally, the judgment module will calculate the characteristic value
  • the multi-weight neural network is input and the quality judgment result is output. According to the output result, it is judged whether the stress treatment of the welding seam to be judged
  • the signal processing and feature extraction module and the judgment module are set in the PC.
  • the sound vibration analyzer sends the digitized acoustic signal to the PC and stores it in the PC.
  • the signal processing and feature extraction module performs filtering and pre-processing on the acoustic signal.
  • the characteristic value is processed and calculated, and then the acquired characteristic value of the acoustic signal is input to the judgment module, and the judgment module outputs the quality judgment result.
  • the method for judging the impact quality of post-weld welds based on intelligent acoustic information recognition extracts features of acoustic signals during ultrasonic stress processing, and uses multi-weight neural network algorithms for pattern recognition. Without damaging the weldment, the stress treatment quality of the welded seam after welding can be quickly and accurately judged, and the cost is low. Greatly improve the reliability of the stress relief process, thereby improving the overall quality of the welding process.
  • Fig. 1 is a flowchart of a method for judging the impact quality of a post-weld weld based on intelligent acoustic information recognition provided by the present invention
  • Figure 2 is a structural diagram of the system for judging the impact quality of post-weld welds based on intelligent acoustic information recognition provided by the present invention
  • FIG. 3 is a structural diagram of the sound signal acquisition hardware platform in the system for judging the impact quality of post-weld weld seam based on intelligent sound information recognition provided by the present invention
  • FIG. 4 is a frequency domain diagram of the sound waveform collected by the sound signal acquisition hardware platform of the present invention before and after the Butterworth filter is filtered.
  • Multi-weight neural networks can achieve the optimal coverage of complex high-dimensional spaces. Samples are used as neuron nodes, and the shortest Euclidean distance is used to construct a hypergeometric body to describe a certain type of neuron, thereby forming a complex training space and realizing classification Recognition function.
  • the multi-weight neural network used in this patent is composed of two geometric bodies in high-dimensional spaces, which represent the qualified and unqualified quality of the post-weld stress treatment respectively. Each geometry is constructed with the shortest Euclidean distance between nodes using training samples as neuron nodes.
  • the mathematical model of the geometry is set with weight parameters and bias parameters, which can fit the best high-dimensional space coverage according to different training samples to achieve optimal identification.
  • a method for judging the impact quality of welds after welding based on intelligent acoustic information recognition which is characterized in that it includes the following steps:
  • the post-weld seam of each processed weldment shall be measured in accordance with the "Indentation Strain Method for the Determination of Residual Stress of Metallic Materials of the People's Republic of China" standard.
  • a resistance strain gauge is used as a sensitive element for measurement, and an indentation is made by impact loading at the center of the strain rose.
  • the strain increase in the elastic area outside the indentation area is recorded by the strain gauge, so as to obtain the true elasticity corresponding to the residual stress. Strain, find the magnitude of the stress;
  • step S3 Establish a multi-weight neural network model, and train the multi-weight neural network model using the acoustic signal sample set marked in step S2) to obtain a multi-weight neural network that can be used to determine the impact quality of the post-weld weld The internet;
  • Step 1 Take the four features of each acoustic signal collected by the training sample collection module as a feature vector.
  • Step 2 Find the minimum value in the N ⁇ N-dimensional matrix A, and the position corresponding to the subscript is the sequence number of the two closest acoustic signal characteristic sample points, which are marked as P 11 and P 12 , and use them The first neuron ⁇ 1 constructed by corresponding two acoustic signal characteristic sample points;
  • Step 3 Delete the points covered by the first neuron ⁇ 1 in the acoustic signal feature sample point set ⁇ A 1 , A 2 ,..., A N ⁇ , and in the remaining acoustic signal feature sample point set, Calculate the distance of each point to point P 11 and point P 12 , find the two points with the shortest distance, record them as P 21 and P 22 , use acoustic signal characteristic sample points P 21 and P 22 to construct the second nerve of MDOFNN Yuan, denoted as ⁇ 2 ;
  • Step 4 follow Step 3, continue to process the remaining acoustic signal characteristic sample points, calculate P i1 P i2 , construct the i-th neuron, denoted as ⁇ i ;
  • the algorithm iteratively obtains the coverage of the two multi-weighted neuron coverage areas of "Qualified Processing Quality” and "Unqualified Processing Quality", and calculates the coverage of the test sample and two multi-weighted neuron networks representing the quality of post-weld weld stress processing.
  • the Euclidean distance between the regions and the Euclidean distance between the "processing quality qualified" multi-weighted neuron coverage area is the type that is the case where the post-weld weld stress processing quality in the test sample is qualified, and the "processing quality”
  • the type of "unqualified" multi-weighted neuron coverage area with a short European distance is the case where the quality of the post-weld weld stress treatment in the test sample is unqualified; after multiple test sample experiments, the multi-weight neural network is verified
  • the recognition accuracy can reach more than 95%.
  • step S4) Obtain the characteristic value of the acoustic signal of the impact treatment of the weld after welding to be judged, and input the characteristic value into the multi-weight neural network trained in step S4), and output the judgment result of the impact treatment quality of the weld to be judged after the weld , That is, it is judged whether the impact treatment of the weld after welding to be judged is "qualified processing quality” or "processing quality unqualified".
  • Step S1) The specific processing process is: controlling the ultrasonic impact gun head to different processing pressure (pressure of the impact gun head relative to the weld toe of the weld), processing speed (speed of the impact gun head relative to the toe of the weld), and processing
  • the angle (the angle of the impact gun head relative to the weld toe of the weld) and the impact frequency (the vibration frequency of the piezoelectric ceramic stack) impact the weld after welding to obtain the acoustic signal during the impact treatment; for example: control the ultrasonic impact gun
  • the processing pressure of the head is 3Kg, 4Kg, 5Kg
  • the processing speed is 16cm/min, 32cm/min
  • the processing angle is 45°, 60°
  • the impact frequency is 60% and 85% of the duty cycle of the controller.
  • the seam welding toe is processed, and the acoustic signal samples under various stress treatment situations manufactured artificially are obtained.
  • the Fourier transform is used to convert the acoustic signal in the time domain into the acoustic signal in the frequency domain.
  • the Butterworth filter is used to filter the acoustic signal in the frequency domain; the Butterworth filter has the largest flat amplitude response, which can effectively remove the high-frequency signal from the acoustic signal in the frequency domain. Its amplitude square function is:
  • N is the order of the filter
  • ⁇ c is the cut-off frequency of the low-pass filter.
  • Butterworth filters have the advantages of balanced characteristics in terms of attenuation slope, linear phase and loading characteristics, which can effectively remove The high-frequency signal in the acoustic signal in the frequency domain.
  • Fig. 4 is a frequency domain diagram of the sound waveform collected in the present invention before and after being filtered by a Butterworth filter.
  • the acoustic signal After filtering the acoustic signal, the acoustic signal is divided into frames to extract its short-term characteristics, and short-time windowing technology is selected for frame processing.
  • the window used is a Hamming window with a window length of 1024 and an overlap of 50%.
  • the short-term zero-crossing rate, short-term average amplitude, short-term energy and short-term zero-energy ratio of the sound signal can be extracted from the time domain;
  • the short-term zero-crossing rate refers to the number of times the signal passes the zero value in a frame of signal; the calculation formula of the short-term zero-crossing rate Z n is as follows:
  • n represents the current sampling time point
  • N is the length of the Hamming window
  • x ⁇ (m) represents the signal after x(m) is windowed
  • x(m) is the amplitude of the acoustic signal at time m value
  • the impact energy of the sound signal is an ultrasonic weld gun impact significant difference between short-term energy E n is calculated as follows:
  • short-term energy is the square of an acoustic signal and in the signal, short-time average magnitude M n is used to calculate the absolute values of the acoustic signal and to measure the variation width; short-term energy calculation formula M n as follows:
  • the short-term zero-energy ratio is the ratio of the zero-crossing rate and the short-term energy within a frame of signal.
  • the short-term zero-energy ratio ZER n is calculated as follows:
  • a post-weld weld impact quality discrimination system based on intelligent acoustic information recognition, including: a sound signal acquisition hardware platform for collecting acoustic signals in the process of post-weld weld impact processing; signal processing and feature extraction modules for matching The acoustic signal is filtered and preprocessed and the eigenvalue is calculated; the discrimination module is used to input the eigenvalue calculated by the signal processing and feature extraction module into the multi-weight neural network and output the quality judgment result.
  • the multi-weight neural network is passed through After training, a multi-weight neural network can be used to judge the impact quality of welds after welding.
  • the sound signal acquisition hardware platform includes: an ultrasonic impact gun 1, a mobile operating platform 2, a weldment to be processed 3, a free-field microphone 4, a sound vibration analyzer 5, and a PC 6, wherein the position of the ultrasonic impact gun 1 is not fixed. Change, the weldment 3 to be processed is fixed on the mobile operating platform 2, and the mobile operating platform 2 can move relative to the ultrasonic impact gun 1 along the length direction of the weldment 3 to be processed. With the mobile operating platform 2 carrying the weldment 3 to be processed The ultrasonic impact gun 1 performs the residual stress treatment on the weld toe of the weld 3 to be processed.
  • the free-field microphone 4 is placed in a circle with the tip of the ultrasonic impact gun 1 as the center and a radius of 1.5m, and is used to collect the analog signal of the sound during the entire processing process and transmit it to the sound and vibration analyzer 5; sound vibration
  • the analyzer 5 converts the received analog signal of the sound into a digitized time-domain sound signal. Then the sound vibration analyzer 5 transmits the received sound information to the PC 6 and the PC 6 stores it in the form of a file.
  • the signal processing and feature extraction module and the judgment module are both set in the PC 6, and then the signal processing and feature extraction module performs filtering and preprocessing on the acoustic signal and calculates the characteristic value; finally, the judgment module inputs the calculated characteristic value into multiple
  • the weighted neural network also outputs the quality judgment result, and judges whether the stress treatment of the weld after welding to be judged is qualified or not according to the output result.
  • the system is used to judge the quality of stress relief treatment, without damaging the weldment, it can quickly and accurately judge the quality of the stress treatment of the weld after welding, and the cost is low.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Biochemistry (AREA)
  • Analytical Chemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Signal Processing (AREA)
  • Acoustics & Sound (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Mechanical Engineering (AREA)
  • Theoretical Computer Science (AREA)
  • Food Science & Technology (AREA)
  • Medicinal Chemistry (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Optics & Photonics (AREA)
  • Quality & Reliability (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)

Abstract

一种基于智能声信息识别的焊后焊缝冲击质量判别方法及系统，包括：控制超声冲击枪（1）枪头以不同的处理压力、处理速度、处理角度和冲击频率对焊后焊缝进行冲击处理，获取冲击处理过程中的声信号，计算声信号的特征值，构建包含多种应力处理情况的声信号样本集；根据焊后焊缝的冲击处理质量测定结果对声信号样本集进行标注；建立多权值神经网络模型，利用经过标注的声信号样本集对多权值神经网络模型进行训练；获取待判别焊后焊缝冲击处理声信号的特征值，将特征值输入经过训练的多权值神经网络，输出待判别焊后焊缝冲击处理质量的判断结果。声音信号采集硬件平台包括：超声冲击枪（1）、移动操作平台（2）、待处理焊件（3）、自由场传声器（4）、声音振动分析仪（5）和PC机（6）。

Description

一种基于智能声信息识别的焊后焊缝冲击质量判别方法及系统 技术领域
本发明涉及机械控制领域，尤其涉及一种基于智能声信息识别的焊后焊缝冲击质量判别方法。
背景技术
焊接作为制造业中不可或缺的重要一环，正在从低功率、低精度的低质量焊接往大功率、高精度的高质量焊接发展，这其中焊接完成后的应力处理在整个高质量焊接过程中就显得尤为重要。焊接完成后的不均匀温度场会导致焊缝中的应力分布不均，从而降低焊件的抗屈服强度、抗疲劳强度，严重会导致焊件形变，焊缝开裂等严重后果。现有的焊缝残余应力的方法中消除质量较好的是超声冲击法，超声冲击技术是一种高效的消除部件表面或焊缝区有害残余拉应力、引进有益压应力的方法。超声冲击设备利用大功率的能量推动冲击头以每秒约2万次的频率冲击金属物体表面，高频、高效和聚焦下的大能量使金属表层产生较大的压缩塑性变形；同时超声冲击改变了原有的应力场，产生有益的压应力；高能量冲击下金属表面温度极速升高又迅速冷却，使作用区表层金属组织发生变化，冲击部位得以强化。
超声冲击法操作过程中的处理速度、压力、角度、钢材种类、厚度等因素都决定了超声冲击法处理效果的好坏，但这些因素难以在操作过程中测量及量化。由于，操作过程中超声冲击声音信号的时频域的特征中包含很多信息，这些声信号特征是操作过程中外界因素的综 合体现，决定了超声冲击的质量。作用在焊件焊缝上的超声冲击质量直接决定了焊缝残余应力消除的程度。因此，亟待寻找一种基于智能声信息识别的焊后焊缝接冲击质量的判别方法。
发明内容
本发明的目的在于提供一种识别准确、监测成本低的基于智能声信息识别的焊后焊缝冲击质量判别方法及系统。
本发明的技术解决方案是：
一种基于智能声信息识别的焊后焊缝冲击质量判别方法，其特征是：包括以下步骤：
S1)控制超声冲击枪枪头以不同的处理压力、处理速度、处理角度和冲击频率对焊后焊缝进行冲击处理，获取冲击处理过程中的声信号，计算声信号的特征值，构建包含多种应力处理情况的声信号样本集；
S2)测定焊后焊缝的冲击处理质量，并根据测定结果对所述声信号样本集进行标注；
测定过程采用电阻应变片作为测量用敏感元件，在应变花中心部位采用冲击加载制造压痕，通过应变仪记录压痕区外弹性区应变增量的变化，从而获得对应于残余应力大小的真实弹性应变，求出应力的大小；
计算应力消除率，应力消除率的计算公式为：应力消除率＝(焊后焊缝冲击处理后的应力/焊后焊缝冲击处理前的应力)*100％；其中，焊后焊缝冲击处理前的应力即为整个焊接件的应力；若应力消除 率高于70％，则标记为冲击质量合格；若所述应力消除率低于70％，则标记为冲击质量不合格；
S3)建立多权值神经网络模型，并利用经过步骤S2)标注得到的声信号样本集对所述多权值神经网络模型进行训练，得到可用于焊后焊缝冲击质量判别的多权值神经网络；
第1步：将训练样本采集模块采集的每个声信号的四个特征作为一个特征向量样本点记作A 1,A 2,…,A N，计算特征样本集合中任意两点之间的距离，并存储在N×N维的矩阵A中，其中A ij表示声信号特征样本点A i到A j的距离，且A ii＝0(i＝1,2,...,N)；
第2步：在N×N维的矩阵A中找到最小值，其下脚标对应的位置即为要找的距离最近的两个声信号特征样本点序号，记为P 11和P 12，并用其对应的两个声信号特征样本点构造的第一个神经元θ 1；
第3步：在声信号特征样本点集合{A 1,A 2,…,A N}中删掉被第一个神经元θ 1覆盖的点，在剩下的声信号特征样本点集合中，计算每个点分别到点P 11和点P 12的距离，找到其中距离最短的两点，记为P 21和P 22，用声信号特征样本点P 21和P 22构造MDOFNN的第二个神经元，记作θ 2；
第4步：遵循第3步，继续对余下的声信号特征样本点进行处理，计算得到P i1P i2，构造第i个神经元，记作θ i；
第5步：当i＝N-1时，说明已经处理完声信号特征样本点集合中的所有点，得到N-1个相连的神经元折线模型；
最终算法迭代获得了“处理质量合格”与“处理质量不合格”这 两个多权值神经元覆盖区域，计算测试样本与代表焊后焊缝应力处理质量的两个多权值神经元网络覆盖区之间的欧式距离，与“处理质量合格”多权值神经元覆盖区域欧式距离较近的那一类即为该测试样本中的焊后焊缝应力处理质量合格的情况，与“处理质量不合格”多权值神经元覆盖区域欧式距离较近的那一类即为该测试样本中的焊后焊缝应力处理质量不合格的情况；
S4)获取待判别焊后焊缝冲击处理声信号的特征值，并将所述特征值输入经过步骤S4)训练得到的多权值神经网络，输出待判别焊后焊缝冲击处理质量的判断结果，即判断待判别焊后焊缝冲击处理为“处理质量合格”，还是“处理质量不合格”。
步骤S1)具体处理过程为：控制超声冲击枪枪头以不同的处理压力、处理速度、处理角度和冲击频率对焊后焊缝进行冲击处理，获取冲击处理过程中的声信号；
使用傅立叶变换将时域上的声信号转化为频域上的声信号，采用巴特沃斯滤波器对频域上的声信号进行滤波；
对声信号进行滤波后，对声信号进行分帧处理来提取它的短时特征，选取短时加窗技术进行分帧处理，其所用的窗口是汉明窗，窗口长度为1024，重叠50％进行分帧；
接着从时域上可以提取声音信号的短时过零率、短时平均幅度、短时能量和短时零能比；
(1)短时过零率特征：
短时过零率是指在一帧信号里信号通过零值的次数；短时过零率 Z n的计算公式如下：
Figure PCTCN2020124189-appb-000001
式中：n代表当前的采样时间点，N为汉明窗的长度，x ω(m)代表x(m)经过加窗处理以后的信号，x(m)是在时间m上声信号的幅值；
(2)短时能量：
在不同的变量下，超声冲击枪冲击焊缝的声音信号的能量有显著的区别，短时能量E n的计算公式如下：
Figure PCTCN2020124189-appb-000002
(3)短时平均幅度：
短时能量是指在一帧信号里的声信号的平方和，短时平均幅度M n是用过计算其绝对值之和来衡量声信号变化幅度；短时能量M n的计算公式如下：
Figure PCTCN2020124189-appb-000003
(4)短时零能比
短时零能比是一帧信号内的过零率和短时能量的比值，短时零能比ZER n计算公式如下：
ZER n＝Z n/E n。
一种基于智能声信息识别的焊后焊缝冲击质量判别方法的专用焊后焊缝冲击质量判别系统，其特征是：包括：声音信号采集硬件平 台，用于采集焊后焊缝冲击处理过程中的声信号；信号处理与特征提取模块，用于对所述声信号进行滤波预处理并计算特征值；判别模块，用于将信号处理与特征提取模块计算得到的特征值输入多权值神经网络并输出质量判别结果，该多权值神经网络为经过训练后可用于判别焊后焊缝冲击质量的多权值神经网络；
所述声音信号采集硬件平台包括：超声冲击枪、移动操作平台、待处理焊件、自由场传声器、声音振动分析仪和PC机；其中，超声冲击枪的位置固定不变，待处理焊件固定在移动操作平台上，移动操作平台可沿待处理焊件长度方向与超声冲击枪1形成相对运动，随着移动操作平台带着待处理焊件的移动，超声冲击枪对待处理焊件焊缝的焊趾进行焊缝残余应力处理；自由场传声器放置在以超声冲击枪的枪头为圆心且半径为1.5m的圆中，用于采集整个处理过程中的声音的模拟信号，并传输到声音和振动分析仪；声音振动分析仪将接受到的声音的模拟信号转化为数字化的时域声信号；然后声音振动分析仪再将接受到的声信息再传送给PC机，由PC机以文件的形式存储起来；信号处理与特征提取模块和判断模块均设置在PC机中，然后由信号处理与特征提取模块对声信号进行滤波预处理并计算特征值；最后，由判别模块将计算得到的特征值输入多权值神经网络并输出质量判别结果，根据输出结果判定待判别焊后焊缝的应力处理是否合格。
信号处理与特征提取模块和判断模块设置在PC机中，声音振动分析仪将数字化的声信号发送至PC机，并存储在PC机中，信号处理与特征提取模块对所述声信号进行滤波预处理并计算特征值，然后将得到 的声信号特征值输入判断模块，由判断模块输出质量判断结果。
与现有技术相比，本发明提供的基于智能声信息识别的焊后焊缝冲击质量判别方法，对超声应力处理过程中的声信号进行特征提取，使用多权值神经网络算法进行模式辨识，无需破坏焊件，便可以快速准确地判断焊后焊缝应力处理质量，成本低廉。极大地提高了应力消除工艺的可靠性，进而提高焊接过程的整体质量。
附图说明
图1为本发明提供的基于智能声信息识别的焊后焊缝冲击质量判别方法流程图；
图2为本发明提供的基于智能声信息识别的焊后焊缝冲击质量判别系统结构图；
图3为本发明提供的基于智能声信息识别的焊后焊缝冲击质量判别系统中声音信号采集硬件平台的结构图；
图4是本发明中声音信号采集硬件平台采集到的声音波形经过巴特沃斯滤波器滤波前后的波形频域图。
具体实施方式
多权值神经网络可以实现对复杂高维空间的最优覆盖，以样本为神经元节点，以最短的欧式距离来构造超几何形体描述某一类神经元，从而构成复杂的训练空间，实现分类识别功能。应用于本专利中的多权值神经网络是由两个高维空间中的几何体构成，分别代表焊后应力处理质量合格与不合格。每一个几何体都是以训练样本作为为神经元节点，以节点间最短的欧式距离来构造。几何体的数学模型设置 了权重参数和偏置参数，可以根据不同的训练样本拟合最佳的高维空间覆盖，实现最优辨识。
一种基于智能声信息识别的焊后焊缝冲击质量判别方法，其特征是：包括以下步骤：
S1)控制超声冲击枪枪头以不同的处理压力、处理速度、处理角度和冲击频率对焊后焊缝进行冲击处理，获取冲击处理过程中的声信号，计算声信号的特征值，构建包含多种应力处理情况的声信号样本集；
S2)测定焊后焊缝的冲击处理质量，并根据测定结果对所述声信号样本集进行标注；
对每一个处理完成的焊件的焊后焊缝按照《中华人民共和国国家金属材料残余应力测定压痕应变法》标准进行测定。测定过程采用电阻应变片作为测量用敏感元件，在应变花中心部位采用冲击加载制造压痕，通过应变仪记录压痕区外弹性区应变增量的变化，从而获得对应于残余应力大小的真实弹性应变，求出应力的大小；
计算应力消除率，应力消除率的计算公式为：应力消除率＝(焊后焊缝冲击处理后的应力/焊后焊缝冲击处理前的应力)*100％；其中，焊后焊缝冲击处理前的应力即为整个焊接件的应力；若应力消除率高于70％，则标记为冲击质量合格；若所述应力消除率低于70％，则标记为冲击质量不合格；
S3)建立多权值神经网络模型，并利用经过步骤S2)标注得到的声信号样本集对所述多权值神经网络模型进行训练，得到可用于焊后 焊缝冲击质量判别的多权值神经网络；
第1步：将训练样本采集模块采集的每个声信号的四个特征作为一个特征向量样本点记作A 1,A 2,…,A N，计算特征样本集合中任意两点之间的距离，并存储在N×N维的矩阵A中，其中A ij表示声信号特征样本点A i到A j的距离，且A ii＝0(i＝1,2,...,N)；
第2步：在N×N维的矩阵A中找到最小值，其下脚标对应的位置即为要找的距离最近的两个声信号特征样本点序号，记为P 11和P 12，并用其对应的两个声信号特征样本点构造的第一个神经元θ 1；
第3步：在声信号特征样本点集合{A 1,A 2,…,A N}中删掉被第一个神经元θ 1覆盖的点，在剩下的声信号特征样本点集合中，计算每个点分别到点P 11和点P 12的距离，找到其中距离最短的两点，记为P 21和P 22，用声信号特征样本点P 21和P 22构造MDOFNN的第二个神经元，记作θ 2；
第4步：遵循第3步，继续对余下的声信号特征样本点进行处理，计算得到P i1P i2，构造第i个神经元，记作θ i；
第5步：当i＝N-1时，说明已经处理完声信号特征样本点集合中的所有点，得到N-1个相连的神经元折线模型；
最终算法迭代获得了“处理质量合格”与“处理质量不合格”这两个多权值神经元覆盖区域，计算测试样本与代表焊后焊缝应力处理质量的两个多权值神经元网络覆盖区之间的欧式距离，与“处理质量合格”多权值神经元覆盖区域欧式距离较近的那一类即为该测试样本中的焊后焊缝应力处理质量合格的情况，与“处理质量不合格”多权 值神经元覆盖区域欧式距离较近的那一类即为该测试样本中的焊后焊缝应力处理质量不合格的情况；经过多个测试样本实验，验证多权值神经网络的识别准确率可以达到95％以上。
S4)获取待判别焊后焊缝冲击处理声信号的特征值，并将所述特征值输入经过步骤S4)训练得到的多权值神经网络，输出待判别焊后焊缝冲击处理质量的判断结果，即判断待判别焊后焊缝冲击处理为“处理质量合格”，还是“处理质量不合格”。
步骤S1)具体处理过程为：控制超声冲击枪枪头以不同的处理压力(冲击枪头相对于焊缝焊趾的压力)、处理速度(冲击枪头相对于焊缝焊趾的速度)、处理角度(冲击枪头相对于焊缝焊趾的角度)和冲击频率(压电陶瓷堆振动频率)对焊后焊缝进行冲击处理，获取冲击处理过程中的声信号；譬如：控制超声冲击枪枪头以处理压力分别为3Kg、4Kg、5Kg，处理速度为16cm/min、32cm/min，处理角度为45°、60°，冲击频率为控制器占空比的60％、85％分别组合对焊缝焊趾部位进行处理，获得人为制造的多种应力处理情况下的声信号样本。
由于实际采集过程中的环境并不是封闭的，即采集的声信号中必然存在干扰噪声，需要对其进行滤波处理，所以使用傅立叶变换将时域上的声信号转化为频域上的声信号，采用巴特沃斯滤波器对频域上的声信号进行滤波；巴特沃斯滤波器具有最大平坦幅度响应，可以有效的去频域上的声信号中的高频信号，其幅值平方函数为：
Figure PCTCN2020124189-appb-000004
式中N为滤波器的阶数，ωc为低通滤波器的截止频率。N值越大，过渡带越抖，通带和阻带的近似性也越好，巴特沃斯滤波器在衰减斜率、线性相位与加载特性三个方面都有特性均衡的优点，可以有效的去频域上的声信号中的高频信号。图4是本发明中采集到的声音波形经过巴特沃斯滤波器滤波前后的波形频域图。
对声信号进行滤波后，对声信号进行分帧处理来提取它的短时特征，选取短时加窗技术进行分帧处理，其所用的窗口是汉明窗，窗口长度为1024，重叠50％进行分帧；
接着从时域上可以提取声音信号的短时过零率、短时平均幅度、短时能量和短时零能比；
(1)短时过零率特征：
短时过零率是指在一帧信号里信号通过零值的次数；短时过零率Z n的计算公式如下：
Figure PCTCN2020124189-appb-000005
式中：n代表当前的采样时间点，N为汉明窗的长度，x ω(m)代表x(m)经过加窗处理以后的信号，x(m)是在时间m上声信号的幅值；
(2)短时能量：
在不同的变量下，超声冲击枪冲击焊缝的声音信号的能量有显著的区别，短时能量E n的计算公式如下：
Figure PCTCN2020124189-appb-000006
(3)短时平均幅度：
短时能量是指在一帧信号里的声信号的平方和，短时平均幅度M n是用过计算其绝对值之和来衡量声信号变化幅度；短时能量M n的计算公式如下：
Figure PCTCN2020124189-appb-000007
(4)短时零能比
短时零能比是一帧信号内的过零率和短时能量的比值，短时零能比ZER n计算公式如下：
ZER n＝Z n/E n。
一种基于智能声信息识别的焊后焊缝冲击质量判别系统，包括：声音信号采集硬件平台，用于采集焊后焊缝冲击处理过程中的声信号；信号处理与特征提取模块，用于对所述声信号进行滤波预处理并计算特征值；判别模块，用于将信号处理与特征提取模块计算得到的特征值输入多权值神经网络并输出质量判别结果，该多权值神经网络为经过训练后可用于判别焊后焊缝冲击质量的多权值神经网络。
所述声音信号采集硬件平台包括：超声冲击枪1、移动操作平台2、待处理焊件3、自由场传声器4、声音振动分析仪5和PC机6，其中，超声冲击枪1的位置固定不变，待处理焊件3固定在移动操作平台2上，移动操作平台2可沿待处理焊件3长度方向与超声冲击枪1形成相对运动，随着移动操作平台2带着待处理焊件3的移动，超声冲击枪1对待处理焊件3焊缝的焊趾进行焊缝残余应力处理。自由 场传声器4放置在以超声冲击枪1的枪头为圆心且半径为1.5m的圆中，用于采集整个处理过程中的声音的模拟信号，并传输到声音和振动分析仪5；声音振动分析仪5将接受到的声音的模拟信号转化为数字化的时域声信号。然后声音振动分析仪5再将接受到的声信息再传送给PC机6，由PC机6以文件的形式存储起来。信号处理与特征提取模块和判断模块均设置在PC机6中，然后由信号处理与特征提取模块对声信号进行滤波预处理并计算特征值；最后，由判别模块将计算得到的特征值输入多权值神经网络并输出质量判别结果，根据输出结果判定待判别焊后焊缝的应力处理是否合格。使用该系统进行应力消除处理质量判断，无需破坏焊件，便可以快速准确地判断焊后焊缝应力处理质量，成本低廉。

Claims (3)

  1. 一种基于智能声信息识别的焊后焊缝冲击质量判别方法，其特征是：包括以下步骤：
    S1)控制超声冲击枪枪头以不同的处理压力、处理速度、处理角度和冲击频率对焊后焊缝进行冲击处理，获取冲击处理过程中的声信号，计算声信号的特征值，构建包含多种应力处理情况的声信号样本集；
    S2)测定焊后焊缝的冲击处理质量，并根据测定结果对所述声信号样本集进行标注；
    测定过程采用电阻应变片作为测量用敏感元件，在应变花中心部位采用冲击加载制造压痕，通过应变仪记录压痕区外弹性区应变增量的变化，从而获得对应于残余应力大小的真实弹性应变，求出应力的大小；
    计算应力消除率，应力消除率的计算公式为：应力消除率＝(焊后焊缝冲击处理后的应力/焊后焊缝冲击处理前的应力)*100％；其中，焊后焊缝冲击处理前的应力即为整个焊接件的应力；若应力消除率高于70％，则标记为冲击质量合格；若所述应力消除率低于70％，则标记为冲击质量不合格；
    S3)建立多权值神经网络模型，并利用经过步骤S2)标注得到的声信号样本集对所述多权值神经网络模型进行训练，得到可用于焊后焊缝冲击质量判别的多权值神经网络；
    第1步：将训练样本采集模块采集的每个声信号的四个特征作为 一个特征向量样本点记作A 1,A 2,…,A N，计算特征样本集合中任意两点之间的距离，并存储在N×N维的矩阵A中，其中A ij表示声信号特征样本点A i到A j的距离，且A ii＝0(i＝1,2,...,N)；
    第2步：在N×N维的矩阵A中找到最小值，其下脚标对应的位置即为要找的距离最近的两个声信号特征样本点序号，记为P 11和P 12，并用其对应的两个声信号特征样本点构造的第一个神经元θ 1；
    第3步：在声信号特征样本点集合{A 1,A 2,…,A N}中删掉被第一个神经元θ 1覆盖的点，在剩下的声信号特征样本点集合中，计算每个点分别到点P 11和点P 12的距离，找到其中距离最短的两点，记为P 21和P 22，用声信号特征样本点P 21和P 22构造MDOFNN的第二个神经元，记作θ 2；
    第4步：遵循第3步，继续对余下的声信号特征样本点进行处理，计算得到P i1P i2，构造第i个神经元，记作θ i；
    第5步：当i＝N-1时，说明已经处理完声信号特征样本点集合中的所有点，得到N-1个相连的神经元折线模型；
    最终算法迭代获得了“处理质量合格”与“处理质量不合格”这两个多权值神经元覆盖区域，计算测试样本与代表焊后焊缝应力处理质量的两个多权值神经元网络覆盖区之间的欧式距离，与“处理质量合格”多权值神经元覆盖区域欧式距离较近的那一类即为该测试样本中的焊后焊缝应力处理质量合格的情况，与“处理质量不合格”多权值神经元覆盖区域欧式距离较近的那一类即为该测试样本中的焊后焊缝应力处理质量不合格的情况；
    S4)获取待判别焊后焊缝冲击处理声信号的特征值，并将所述特征值输入经过步骤S4)训练得到的多权值神经网络，输出待判别焊后焊缝冲击处理质量的判断结果，即判断待判别焊后焊缝冲击处理为“处理质量合格”，还是“处理质量不合格”。
  2. 根据权利要求1所述的基于智能声信息识别的焊后焊缝冲击质量判别方法，其特征是：步骤S1)具体处理过程为：控制超声冲击枪枪头以不同的处理压力、处理速度、处理角度和冲击频率对焊后焊缝进行冲击处理，获取冲击处理过程中的声信号；
    使用傅立叶变换将时域上的声信号转化为频域上的声信号，采用巴特沃斯滤波器对频域上的声信号进行滤波；
    对声信号进行滤波后，对声信号进行分帧处理来提取它的短时特征，选取短时加窗技术进行分帧处理，其所用的窗口是汉明窗，窗口长度为1024，重叠50％进行分帧；
    接着从时域上可以提取声音信号的短时过零率、短时平均幅度、短时能量和短时零能比；
    (1)短时过零率特征：
    短时过零率是指在一帧信号里信号通过零值的次数；短时过零率Z n的计算公式如下：
    Figure PCTCN2020124189-appb-100001
    式中：n代表当前的采样时间点，N为汉明窗的长度，x ω(m)代表x(m)经过加窗处理以后的信号，x(m)是在时间m上声信号的幅值；
    (2)短时能量：
    在不同的变量下，超声冲击枪冲击焊缝的声音信号的能量有显著的区别，短时能量E n的计算公式如下：
    Figure PCTCN2020124189-appb-100002
    (3)短时平均幅度：
    短时能量是指在一帧信号里的声信号的平方和，短时平均幅度M n是用过计算其绝对值之和来衡量声信号变化幅度；短时能量M n的计算公式如下：
    Figure PCTCN2020124189-appb-100003
    (4)短时零能比
    短时零能比是一帧信号内的过零率和短时能量的比值，短时零能比ZER n计算公式如下：
    ZER n＝Z n/E n。
  3. 一种权利要求1所述的基于智能声信息识别的焊后焊缝冲击质量判别方法的专用焊后焊缝冲击质量判别系统，其特征是：包括：声音信号采集硬件平台，用于采集焊后焊缝冲击处理过程中的声信号；信号处理与特征提取模块，用于对所述声信号进行滤波预处理并计算特征值；判别模块，用于将信号处理与特征提取模块计算得到的特征值输入多权值神经网络并输出质量判别结果，该多权值神经网络为经过训练后可用于判别焊后焊缝冲击质量的多权值神经网络；
    所述声音信号采集硬件平台包括：超声冲击枪、移动操作平台、待处理焊件、自由场传声器、声音振动分析仪和PC机；其中，超声冲击枪的位置固定不变，待处理焊件固定在移动操作平台上，移动操作平台可沿待处理焊件长度方向与超声冲击枪1形成相对运动，随着移动操作平台带着待处理焊件的移动，超声冲击枪对待处理焊件焊缝的焊趾进行焊缝残余应力处理；自由场传声器放置在以超声冲击枪的枪头为圆心且半径为1.5m的圆中，用于采集整个处理过程中的声音的模拟信号，并传输到声音和振动分析仪；声音振动分析仪将接受到的声音的模拟信号转化为数字化的时域声信号；然后声音振动分析仪再将接受到的声信息再传送给PC机，由PC机以文件的形式存储起来；信号处理与特征提取模块和判断模块均设置在PC机中，然后由信号处理与特征提取模块对声信号进行滤波预处理并计算特征值；最后，由判别模块将计算得到的特征值输入多权值神经网络并输出质量判别结果，根据输出结果判定待判别焊后焊缝的应力处理是否合格。
PCT/CN2020/124189 2019-11-08 2020-10-28 一种基于智能声信息识别的焊后焊缝冲击质量判别方法及系统 WO2021068983A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/288,560 US11927563B2 (en) 2019-11-08 2020-10-28 Smart acoustic information recognition-based welded weld impact quality determination method and system

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911088740.4A CN110824006B (zh) 2019-11-08 2019-11-08 一种基于智能声信息识别的焊后焊缝冲击质量判别方法
CN201911088740.4 2019-11-08

Publications (1)

Publication Number Publication Date
WO2021068983A1 true WO2021068983A1 (zh) 2021-04-15

Family

ID=69553566

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/124189 WO2021068983A1 (zh) 2019-11-08 2020-10-28 一种基于智能声信息识别的焊后焊缝冲击质量判别方法及系统

Country Status (3)

Country Link
US (1) US11927563B2 (zh)
CN (1) CN110824006B (zh)
WO (1) WO2021068983A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117283094A (zh) * 2023-11-22 2023-12-26 内蒙古工业大学 一种可自动追踪以及施加超声辅助的焊接系统
CN117433666A (zh) * 2023-10-20 2024-01-23 江苏省特种设备安全监督检验研究院 基于应力检测的压力容器故障预警方法及系统

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110824006B (zh) * 2019-11-08 2021-12-28 南通大学 一种基于智能声信息识别的焊后焊缝冲击质量判别方法
CN113049683B (zh) * 2021-03-10 2022-05-10 南通大学 一种面向水下管线的超声波探伤装置
CN115418472A (zh) * 2022-08-04 2022-12-02 上海海事大学 一种超声振动时效系统
CN116698977B (zh) * 2023-08-07 2023-10-31 山东科技大学 一种基于声音信号的煤矸识别传感器及识别方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101492759A (zh) * 2009-03-03 2009-07-29 南京工业大学 一种抗应力腐蚀开裂的超声冲击处理工艺
US8146429B2 (en) * 2009-08-03 2012-04-03 Georgia Tech Research Corporation Methods and systems for classifying the type and severity of defects in welds
CN102489859A (zh) * 2011-12-16 2012-06-13 长春三友汽车部件制造有限公司 一种提高先进高强钢点焊接头力学性能的方法
CN105171332A (zh) * 2015-11-02 2015-12-23 哈尔滨电机厂有限责任公司 贯流式水轮发电机转子支架焊缝裂纹的处理方法
WO2016062897A1 (en) * 2014-10-24 2016-04-28 Renishaw Plc Acoustic apparatus and method for inspection of an object
CN109101987A (zh) * 2017-06-20 2018-12-28 林肯环球股份有限公司 用于焊接件分类和关联的机器学习
CN110824006A (zh) * 2019-11-08 2020-02-21 南通大学 一种基于智能声信息识别的焊后焊缝冲击质量判别方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105637338A (zh) * 2013-11-05 2016-06-01 米其林集团总公司 用于轮胎异常的非破坏性检测的方法和装置
CN104102923A (zh) * 2014-07-16 2014-10-15 西安建筑科技大学 一种基于mfcc算法的朱鹮个体识别方法
CN106992011A (zh) * 2017-01-25 2017-07-28 杭州电子科技大学 基于mf‑plpcc特征的工程机械声音识别方法
CN107045718A (zh) * 2017-03-24 2017-08-15 南通大学 焊后焊缝三维追踪及应力处理方法
CN107545890A (zh) * 2017-08-31 2018-01-05 桂林电子科技大学 一种声音事件识别方法
CN109258509B (zh) * 2018-11-16 2023-05-02 太原理工大学 一种生猪异常声音智能监测系统与方法
CN109350014A (zh) * 2018-12-10 2019-02-19 苏州小蓝医疗科技有限公司 一种鼾声识别方法与系统
CN109448726A (zh) * 2019-01-14 2019-03-08 李庆湧 一种语音控制准确率的调整方法及系统

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101492759A (zh) * 2009-03-03 2009-07-29 南京工业大学 一种抗应力腐蚀开裂的超声冲击处理工艺
US8146429B2 (en) * 2009-08-03 2012-04-03 Georgia Tech Research Corporation Methods and systems for classifying the type and severity of defects in welds
CN102489859A (zh) * 2011-12-16 2012-06-13 长春三友汽车部件制造有限公司 一种提高先进高强钢点焊接头力学性能的方法
WO2016062897A1 (en) * 2014-10-24 2016-04-28 Renishaw Plc Acoustic apparatus and method for inspection of an object
CN105171332A (zh) * 2015-11-02 2015-12-23 哈尔滨电机厂有限责任公司 贯流式水轮发电机转子支架焊缝裂纹的处理方法
CN109101987A (zh) * 2017-06-20 2018-12-28 林肯环球股份有限公司 用于焊接件分类和关联的机器学习
CN110824006A (zh) * 2019-11-08 2020-02-21 南通大学 一种基于智能声信息识别的焊后焊缝冲击质量判别方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117433666A (zh) * 2023-10-20 2024-01-23 江苏省特种设备安全监督检验研究院 基于应力检测的压力容器故障预警方法及系统
CN117433666B (zh) * 2023-10-20 2024-05-07 江苏省特种设备安全监督检验研究院 基于应力检测的压力容器故障预警方法及系统
CN117283094A (zh) * 2023-11-22 2023-12-26 内蒙古工业大学 一种可自动追踪以及施加超声辅助的焊接系统
CN117283094B (zh) * 2023-11-22 2024-01-26 内蒙古工业大学 一种可自动追踪以及施加超声辅助的焊接系统

Also Published As

Publication number Publication date
US11927563B2 (en) 2024-03-12
US20220276204A1 (en) 2022-09-01
CN110824006A (zh) 2020-02-21
CN110824006B (zh) 2021-12-28

Similar Documents

Publication Publication Date Title
WO2021068983A1 (zh) 一种基于智能声信息识别的焊后焊缝冲击质量判别方法及系统
US20210287098A1 (en) On-line prediction method of surface roughness of parts based on sdae-dbn algorithm
Lv et al. Automated control of welding penetration based on audio sensing technology
CN109142547B (zh) 一种基于卷积神经网络的声学在线无损检测方法
CN102528225B (zh) 一种gtaw焊接熔透状态的声音信号传感及预测方法
Li et al. A novel ensemble deep learning model for cutting tool wear monitoring using audio sensors
CN104259081B (zh) 一种压电换能器频率跟踪方法及系统
CN109855874B (zh) 一种声音辅助振动微弱信号增强检测的随机共振滤波器
CN106841403A (zh) 一种基于神经网络的声学玻璃缺陷检测方法
CN103776903B (zh) 一种风电叶片脱层检测方法及检测系统
CN111537610A (zh) 一种用于金属曲板损伤定位的传感器阵列优化方法
Ruan et al. Use AF-CNN for end-to-end fiber vibration signal recognition
CN115128130B (zh) 一种基于动态电阻信号的电阻点焊质量在线评估系统和方法
CN108009514B (zh) 球磨机料位预测方法
CN108008262B (zh) 一种局部放电快速检测方法
Zhao et al. Damage location detection of the CFRP composite plate based on neural network regression
CN115754010A (zh) 机械密封环密封端面质量控制方法
CN114663425A (zh) 基于生成对抗网络的超声焊接质量在线监测方法
Bessonova et al. Investigation of spatial averaging effect of membrane hydrophones for working frequencies in the low MHz range
CN106840281A (zh) 一种基于类方波前馈控制随机共振的涡街频率检测方法
Chen et al. Study on fault diagnosis of ultra-low-speed rolling bearings based on full vector sound spectrogram
TWI793626B (zh) 結構劣化的聲振診斷方法與系統
CN108918994A (zh) 一种自适应拟合提升小波包降噪分析的电能质量扰动检测方法
CN116136441B (zh) 基于非线性Lamb波的螺栓早期松动预紧状态监测方法、系统及装置
CN113113052B (zh) 一种离散点的语音基音识别装置及计算机存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20874176

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20874176

Country of ref document: EP

Kind code of ref document: A1