CN100341658C - High-energy beam welding process multi-signal fusion-monitoring instrument - Google Patents
High-energy beam welding process multi-signal fusion-monitoring instrument Download PDFInfo
- Publication number
- CN100341658C CN100341658C CNB200510080198XA CN200510080198A CN100341658C CN 100341658 C CN100341658 C CN 100341658C CN B200510080198X A CNB200510080198X A CN B200510080198XA CN 200510080198 A CN200510080198 A CN 200510080198A CN 100341658 C CN100341658 C CN 100341658C
- Authority
- CN
- China
- Prior art keywords
- signal
- data
- welding
- monitoring
- fusion
- Prior art date
- Legal status (The legal status 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 status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The present invention relates to an instrument for real-time monitoring of the welding process of high-energy density beam flows, such as laser, electron beams, plasma welding, etc. by combining multiple signals. The present invention is composed of a signal real-time monitoring system, a welding data analysis system and a remote supervision system, wherein the signal real-time monitoring system comprises four single monitoring units, namely a sound signal unit, a visual light unit, an infrared light unit and a CCD video camera unit, and the four single monitoring units are respectively used for collecting light emitted by plasma bodies, sound signals and infrared radiation signals of a molten pool in the process of laser welding; four groups of sensors are simultaneously used to quantitatively measure ultraviolet light, visual light, infrared light, audible sound and ultrasound in the process of high-energy density beam welding, an A/D collection card can simultaneously collect signals of the sound signal unit, the visual light unit and the infrared light unit, and the of the sound signal unit, the visual light unit and the infrared light unit are simultaneously transmitted into a data server and are combined with data stored in DSP. Besides, picture signals of the CCD video camera unit are collected by a picture collection card, and the picture signals are simultaneously transmitted to the data server and are combined with the data of the DSP; the instrument carries out remote monitoring through local area networks.
Description
Technical field: the present invention relates to a kind of monitor that adopts the many signal fused of high energy beam current welding process that many signal fused monitor in real time to high energy beam fluid welding termination process such as laser, electron beam, plasma weldings.
Background technology: many physics and chemical phenomenons such as meeting generation sound, light, electricity, heat in the high energy beam current welding process, these phenomenons have all reflected the quality state information of welding process to a certain extent.The signal that detects and analyze various physical chemical phenomenons is the important means of monitoring welding process always.The signal of main research has sound, infrared light, royal purple light, acoustic emission, the signal of telecommunication etc.Early stage research all is to choose a certain signal as detecting parameter welding process to be monitored in real time, there is the scholar to utilize the method for measuring multiple signal and merging to carry out the real-time monitoring of Laser Welding Quality in recent years, the photic plasma sound of the favourable usefulness of PROMETEC company as Germany, the instrument of optical monitoring signal laser beam welding, Michigan university, Britain's institute of welding has all carried out the research work of this aspect, usually choose ultraviolet light, two kinds of signals in the infrared and audible sound carry out signal fused, though compare mono signal, its reliability all has reinforcement, but, to the welding source, welding material and welding procedure have bigger dependence, can only be to the specific weld source, welding material and technology are monitored; Dynamic process for the plasma variations in the molten bath in the welding process can not be monitored in real time.
Summary of the invention: the monitoring method that the purpose of this invention is to provide a kind of many signal fused of high energy beam current welding process that can adapt to more welding source, welding material and welding procedure and can monitor in real time the change procedure in the molten bath.Technical solution of the present invention is that monitoring system is divided into three parts: real time monitoring signals system, welding data analytical system and remote supervision system.Wherein the real time monitoring signals system mainly finishes real-time detection, real-time analysis, the demonstration in real time to laser beam welding medium blue purple light signal, acoustic signals and infrared signal, welding parameter and welding process data can be saved in file system, also provide simultaneously at the data base administration of welding parameter and the self-checking function of signal condition module and data acquisition module; The welding data analytical system is mainly finished the off-line analysis and the data analysis report generation function of welding process data; The real time monitoring signals system is connected by file storage mechanism with the welding data analytical system, and the real time monitoring signals system is connected by the network data services device with remote supervision system.The real time monitoring signals system comprises four road independent monitoring means: acoustical signal unit, royal purple light unit, infrared light unit and ccd video camera unit, gather the infrared radiation signal in light, acoustical signal and the molten bath of the plasma emission in the laser beam welding, utilize four groups of sensors simultaneously the high energy beam current welding process to be carried out quantitative measurment, these four groups of sensors are respectively: royal purple light optical pickocff, CCD optical pickocff, infrared light transducer[sensor, sonic transducer (audible sound and ultrasonic).Wherein the signal of royal purple light optical pickocff, infrared light transducer[sensor and No. one sonic transducer signal are transported to the A/D capture card through signal conditioning circuit amplification and filtering respectively, the A/D capture card is gathered the signal of sound wave, royal purple light, three unit of infrared light simultaneously, sends in the lump and carries out data fusion in the data server; The picture signal of the CCD optical pickocff in the ccd video camera unit is imported another data server through image pick-up card, carries out remote monitoring by LAN.The acoustical signal unit comprises audio signal and ultrasonic signal, and the condenser type sonic transducer detects the characteristic signal of part of the audible sound below the 20KHz and the part of the ultrasonic wave more than the 20KHz in 4～100kHz scope simultaneously.Its processing method of data comprises starting device and carries out the step of equipment self-inspection; Be used for the step that the user is provided with the monitoring welding parameter; Be used for the step of monitoring in real time; The step that is used for user interface; Be used for the step that signal and data analysis are handled; The step that is used for report generation; The step that is used for network monitor; It is characterized in that the processing method of monitoring comprises in real time:
Be used for step to the welding real-time data acquisition;
Be used to comprise to the statistical analysis of data and the processing of frequency-domain analysis and data characteristics extraction and data fusion;
Carry out the extraction of data characteristics according to the result of statistical analysis and frequency-domain analysis, these features comprise: the Estimation of Mean of royal purple light, infrared signal, the variance of acoustic signals estimates that three road signals are at the frequency band of 2 ~ 3KHz, and acoustic signals is at the frequency band of 15～20KHz and 30～35kHz;
Be used for above-mentioned characteristic signal is merged the step of calculating, this step can merge calculating with a kind of in following two kinds of algorithms: (1) based on the artificial neural network algorithm of back-propagation algorithm, network structure is the structure of 8-n-1; (2) algorithm of SVMs, the kernel function of employing has linear function, polynomial function, RBF and sigmoid function;
Generate the step of following information: the unstable Realtime Alerts of welding process, poor weld statistical analysis, weld defect size and position indicating and typical penetration state are judged.
Opto-collection system of the present invention utilizes optical filter and condenser lens to constitute the attenuated optical signal device, and optical signal adopts the off-axis acquisition mode, and mating plate and collector lens converge on the photodiode photosurface after filtration; Optical filter and collector lens are isolated between photoelectric sensor and solder joint; the high light that produces during on the one hand to welding is decayed; prevented that the photoelectric sensor output current is saturated; protect photoelectric sensor, also made the optical signal of specific wavelength can waltz through light path system on the other hand.Optical filter and condenser lens are installed in inwall and get on the bus in the cylinder of helix groove, can realize that the stepless continuous of image distance is regulated.Adopt Si photodiode and InGaAs PIN photodiode respectively as the sensor of royal purple light and infrared light, make that the linearity of sensor is good, photosensitive area is big, the difficulty that light path is aimed at when having reduced the laser weld monitoring.
The present invention measures and analyzes the acoustic signals of 4～100kHz of collecting, not only can be in the monitoring of 20kHz with interior plasma sound wave signal, and can in 20kHz～100kHz wave band, measure and analyze the variation of misalignment, defocusing amount etc., more reflect the change procedure of inside, molten bath.
The present invention adopts at a high speed, multichannel PowerDAQ series capture card, and this data collecting card has (DSP) process chip of digital signal processor, can realize the quick A/D conversion of multichannel analog data; This collection clamp carries the FIFO memory of 16KB and adopts the pci interface mode simultaneously; Can realize the quick transmission of data.In addition, this capture card is supported third party softwares such as LabWindows/CVI, LabView, and powerful driver is provided, and makes that the function of this monitoring system is abundanter, and the self-checking function at data collecting card can be provided.
The software demarcation of system is seven modules: data acquisition module, subscriber interface module, signal and data analysis processing module, report generation module, System self-test module, real-time monitoring modular, network monitoring module.Software section is realized extraction, the real-time monitoring of welding process stability and the identification of typical penetration state of characteristic signal by powerful signal analysis means, effectively utilize digital signal processor (DSP) and realized the real-time monitoring of many signal fused, and utilize ultrasonic signal to realize the monitoring analysis of blast of article on plasma body and molten bath several data.
The characteristics of comprehensive existing various data fusion system structures have designed a kind of structure of new data fusion system according to the actual conditions of this project, and the multi-sensor data-fusion system that is adopted is constructed as follows shown in the figure.The data that collect from a plurality of sensors are at first passed through preliminary treatments such as the amplification, filtering of signal conditioning circuit, carry out feature extraction then, enter the Data Fusion module at last and obtain the data fusion result.This structure biggest advantage is exactly that Data Fusion was divided for two stages---feature extraction phases and data anastomosing algorithm stage.Mainly solve how from the initial data that sensor collects, to find the characteristic that can reflect the actual welding process feature in feature extraction phases.The present invention adopts traditional feature extracting method based on physical model, and the process of feature extraction depends on the result of early stage to original data signal analysis and processing.In the welding experiment in early stage, carry out statistical analysis and frequency-domain analysis (comprising short time discrete Fourier transform, Gabor spectral transformation, wavelet transformation) through the initial data of obtaining, then analyzing and the welding result of data result that data are handled and reality compares, can determine the Partial Feature of the royal purple optical signal, acoustic signals and the infrared signal that collect from sensor.Wherein apparent in view feature has the Estimation of Mean of royal purple light, infrared signal, the variance of acoustic signals is estimated, and three the road signal have characteristic frequency at the frequency band of 2～3KHz, there is characteristic frequency in acoustic signals at the frequency band of 15～20KHz and 30～35KHz, can provide 8 features to use for data anastomosing algorithm altogether.Different characteristic signal is not quite similar to the expressive ability of weld defect in different welding processes.And how mainly to pay close attention to the analytical characteristic data result of realistic welding in the data anastomosing algorithm stage, the present invention adopts the characteristic signal classification of carrying out different welding result based on the machine learning algorithm of sample, its main method has two: first method is based on the artificial neural network of back-propagation algorithm, network structure is the structure of 8-n-1, be 8 nodes of input layer, the individual neuron of hidden layer n (4≤n≤13), 1 node of output layer can be imported each sample (8 dimension column vectors) and accurately be divided into penetration state and penetration state not.Second method is SVMs, and the kernel function of employing has linear function, polynomial function, RBF and sigmoid function.So just feature extracting method and data anastomosing algorithm can be carried out modularized design respectively, arrange in pairs or groups at different feature extracting methods and data anastomosing algorithm, so just can be at different welding processes, from 8 characteristic signals that characteristic extracting module provides, choose flexibly combination arbitrarily send into two kinds of blending algorithms in the data fusion module in any one analyze.Can search out optimum composition method by test of many times to the actual welding process at different welding material and technology.For example, in the application process of reality, if select 8 whole features, whole system can access maximum defect recognition accuracy, but feature selecting must be many more, and it is also many more that data fusion is calculated the time that is consumed.Require very high occasion in real-time, can only select the Estimation of Mean of royal purple light, infrared signal and the variance of acoustic signals to estimate this 3 features, at this moment system can reach maximum arithmetic speed, but the defect recognition accuracy also declines to a great extent thereupon.In order to obtain the balance of computational speed and recognition correct rate, can select the Estimation of Mean of royal purple light, infrared signal and acoustic signals these 5 features of frequency band, can in short operation time, obtain higher defect recognition accuracy at 2～3kHz, 15～20kHz and 30～35kHz.Adopt short time discrete Fourier transform, statistic decision method, D-S evidential reasoning method respectively and carry out data computation based on the algorithm of artificial neural network.In addition, owing between the characteristic extracting module of the output of sensor and data fusion, added the signal condition module section, make the data anastomosing algorithm module can access the parameter (as signal gain, filter bandwidht etc.) that the signal condition module is adopted, and under situation about handling in real time, data fusion module can be provided with the signal condition parameter according to actual conditions such as input data and fusion results, make the signal of input feature vector extraction module always be in the best effort interval, to obtain fusion results more accurately.
Description of drawings:
Fig. 1 is a hardware system block diagram of the present invention;
Fig. 2 is a software systems block diagram of the present invention;
Fig. 3 is a photodiode work pattern schematic diagram of the present invention;
Fig. 4 is visible light sensor circuit theory diagrams of the present invention;
Fig. 5 is infrared light transducer[sensor circuit theory diagrams of the present invention;
Fig. 6 is voice signal modulate circuit figure of the present invention;
Fig. 7 is a light path part signal conditioning circuit block diagram of the present invention;
Fig. 8 is an index path of the present invention;
Fig. 9 is a flow chart of the present invention.
The specific embodiment:
Hardware of the present invention: one, 4 sensors: 1, sound transducer adopts the Denmark must triumphant (Br ü el﹠amp; Kj r) the free field 1/4-inch of company microphone 4349-A-011; can carry out high sound pressure and high-frequency measurement; owing to adopted stainless steel and protection grid, go for the measurement under adverse circumstances, can satisfy the needs that native system is used in the complex industrial environment.Because this sensor has the frequency response that is higher than 20kHz, therefore except measurement, can also carry out Measurement and analysis to ultrasonic signal as audible sound.2, the royal purple optical sensor adopts the Japanese Si of a company photodiode product, its inner integrated prepositionly puts in advance.Silicon photoelectric diode and amplifying unit are integrated on the chip of a silicon, come the closed loop gain of control amplifier by feedback resistance.Because integrated design, this device can suppress noise and interference preferably, and shell also has shielding action, and the whole sensor structure is small and exquisite, has only 13.2 * 7.32mm
2Put in advance have very little bias current (maximum 64pA) thus make that feedback resistance can be higher.The photosensitive area of photoelectric diode is also very big, the difficulty that light path is aimed at when having reduced the laser weld monitoring.And it has sensitivity preferably to the visible-range wavelength, the visible light that relatively more suitable monitoring of plasma sends.Therefore this diode needn't consider to design voluntarily pre-amplification circuit owing to built-in put in advance, and itself output is exactly voltage signal.And can be according to the intensity variations situation, the measurement category that the voltage signal of selecting suitable feedback resistance to make to be exported is allowed at capture card.According to the photoelectric tube operation principle, be chosen as photovoltaic mode at this, circuit theory diagrams are shown in figure [5].[operation principle of photodetector is mainly based on the photoelectric effect that interaction produced and the fuel factor of light radiation and material.Photoelectric effect comes down to be bound by in incident light and the material interaction of the electronics or the free electron of lattice.According to whether launching electronics, photoelectric effect is divided into inner photoeffect and external photoeffect.Inner photoeffect comprises photoconductive effect, photovoltaic effect, photon drag effective and photoelectromagnetic effect etc. again.
3, infrared light transducer[sensor adopts the Japanese InGaAs PIN of a company type photodiode product, and it has very big shunt resistance (100M Ω) and very little noise.Adopt quartz window, the T0-18 encapsulation, and also its photosensitive area is very big, has reached Φ 5mm, and so can be when making light path system because of the little deviation indeterminacy not occurring.
4, (the used CCD model of system that the model of this system, the factory of product etc. please accuse is the CCD camera system: CV-M77, manufacturer: Denmark The Mechademic Company).
Two, signal conditioning circuit: often amplitude is very little from the signal of sensor output, but also can be subjected to the extraneous interference of electromagnetic field, and in order to obtain being suitable for the voltage signal that capture card is gathered, it is just essential therefore to be used for modulate circuit that signal is handled.Signal conditioning circuit comprises optical signal modulate circuit harmony signal conditioning circuit, and signal conditioning circuit has pre-amplification circuit, low-pass filter circuit, programmable amplifying circuit.Signal conditioning circuit can be used block diagram 6, Fig. 7 (in the dotted line) expression.1, the optical signal modulate circuit comprises pre-amplification circuit, amplifying circuit and filter circuit, because the royal purple optical sensor has had inner preposition amplifier section, preamplifier is set no longer in addition;
Infrared light transducer[sensor in order to obtain voltage signal, need level connect an I-V change-over circuit (preamplifier) owing to have only an independent photodiode behind photoelectric tube.Linearity when measuring for improving silicon photoelectric diode must proof load impedance convergence zero.
In the laser beam welding,, also make the photoelectric tube output current that the variation of tens times even hundreds of times takes place because of materials to be welded change of properties or welding condition change the fluctuation that all will cause optical signal (comprising visible light and infrared light) big amplitude to occur.Saturated in order to prevent that output from appearring in preamplifier, it is excessive that the feedback resistance in the preamplifier can not be chosen.Choosing by testing of feedback resistance is definite, and feedback resistance chose the maximum output voltage value that will make the output of preamplifier just be lower than amplifier slightly when both the light intensity that occurs when laser weld was the strongest.After feedback resistance is determined, if when faint light occurring in the welding process, the prestage output voltage signal will be very low, be subjected to outside noise easily and disturb, and be unfavorable for that also long-distance transmissions arrives capture card, therefore be necessary to add after prestage the amplification circuit again.Select the programme-controlled instrument amplifier PGA202/203 of U.S. BURR-BROWN company among the present invention for use, it need not peripheral chip, and PGA202 and PGA203 cascade are used 16 kinds of programme-controlled gains can forming from 1-8000 times, uses quite convenient, flexible.Select for use eight rank active filters continuous time of U.S. MAXIM company to finish the filtering of optical signal simultaneously.
In the actual welding process, very significantly changing may appear in light signal strength, and the prestage output voltage also will alter a great deal, and in the data acquisition data handling system, the general requirements system is within the range of linearity, and enough precision are arranged.So when the change in electric scope that obtains from sensor or detector is very big, adopt the voltage or the current amplifier of fixed gain can not satisfy above-mentioned two basic demands.When signal hour, for guaranteeing the accuracy of reading or the analog-to-digital precision of indicating meter, wish that amplifier has sufficiently high gain, make meter reading be in high position data output near full journey value or analog-to-digital conversion.When but high-gain amplifier is imported large-signal, amplifier will occur the overload and saturated.On the contrary, as satisfy large-signal input and do not make the amplifier overload just require amplifier to reduce its gain, when running into the small-signal input, the amplifier output valve is too little, makes indicating meter reading or A/D conversion accuracy not enough, causes the measuring accuracy of whole system to reduce.Therefore running into input range when bigger, the requirement signal amplifier can be according to the size of input signal, utilize certain procedure, automatically change its gain, the output voltage of amplifier is remained near within the scope of full scale value, and this amplifier is referred to as gain-programmed amplifier.
Acoustical signal conditioning amplification module adopts Denmark Bi Kai company auxiliary products, comprises with probe connects as one putting 2670 and condition amplification module 2690-OS4 in advance.Put 2670 amplifications that are used for 1/4-inch and 1/8-inch sonic transducer output signal in advance, the 1/4-inch microphone can directly be connected with 2670.Condition amplifier 2690A-OS4 is a kind of four-way condition amplifier, has the characteristics of low noise, the big responding range of multichannel, and it can be configured to different input patterns and port number according to user's needs.Can work in DC10V-33V, also can power by internal battery.
Conditioning module gain control and selftest module are decided to be 16 grades according to actual operating position, can adopt program Automatic Styles or manual set-up mode that the gain amplifier of the signal condition module of royal purple optical signal and infrared signal is set.When carrying out the System self-test survey, this module also can be imported standard measuring signal the signal condition modular circuit of royal purple optical signal and infrared signal, and cooperates the data acquisition software module that the signal condition module is carried out self check.
Three, the core of data acquisition module---A/D data collecting card has been selected the PowerDAQ PD2-MFS-4-500/14 type capture card of U.S. UEI company (United Electronic Industries) for use.The synchronous data collection card of this pci interface has the DSP process chip, can realize the quick collection of multichannel analog data.
Module is chosen and is finished, and carries out overall system design, and whole defect inspection system comprises four road independent monitoring means: acoustical signal unit, visible light unit, infrared light unit and ccd video camera unit.Wherein first three road enters the A/D capture card after all being converted to voltage signal at last, and the A/D capture card of choosing has the function that three road signals are gathered simultaneously, can satisfy real-time undistorted collection requirement.Comprise polarity switching, programmable amplifying circuit, low-pass filter circuit and output buffer in the signal condition module of visible light part, comprise programmable amplifying circuit, low-pass filter circuit and output buffer in the signal condition module of infrared light part.Because royal purple optical sensor output signal is a negative value, therefore need add a polarity switching; The gain adjustment end of programmable amplifying circuit is controlled by the digital output of capture card, so signal condition module and A/D capture card are bidirectional data transfers; Output buffer is in order to improve driving force, to make the signal condition module undistorted after longer Distance Transmission to the signal of capture card.
Because the signal that collects all is to characterize same defect state, in order to make full use of information, improves reliability, stability and the reaction speed of the judgement of system, has used multi-sensor information fusion technology.It is to make full use of a plurality of sensor resources, by reasonable domination and use to these sensors and observation information thereof, come a plurality of sensors in the space or temporal redundancy or complementary information make up according to certain criterion, explain or describe with the uniformity that obtains measurand, make this sensor-based system obtain the more superior performance of the system that subclass constituted therefrom than its each part.
Opto-collection system can behind mating plate, the convex lens converge to visible light in the laser beam welding and infrared light on the photodiode photosurface after filtration.Optical filter and condenser lens constitute the attenuated optical signal device, and they are isolated between photoelectric sensor and solder joint, and this high light that produces when butt welding connects is on the one hand decayed, and has prevented that the photoelectric sensor output current is saturated, has protected photoelectric sensor.Also making the light of specific wavelength pass through light path system on the other hand, is the royal purple light of wave-length coverage 400-440nm for visible light, is that centre wavelength is 1064nm for infrared light, and half-breadth is the optical band of 10nm.The schematic diagram of opto-collection system as shown in Figure 8.
At the tube inwall helix groove mobile lens of having got on the bus of assembling said elements, realize that the stepless continuous of image distance is regulated.During fixed light electric transducer position, use for reference the photograph microstructure, the detachable apparatus that has adopted shell fragment to compress is observed image position with frosted glass earlier, is changed to photoelectric sensor again after regulating image distance.
Software configuration framework of the present invention:
Data acquisition module
The function of data acquisition module can be divided into following components on realizing:
1) multi channel signals high speed acquisition
Comprising: the plasma resonance signals collecting (260nm～700nm); Molten bath infrared radiation signal collection (900nm～1600nm); Acoustical signal collection (can be listened and supersonic range 4～100kHz); Aperture and Molten pool image gathering.
2) the signals collecting data are filed automatically
During signals collecting, the initial data of collection is deposited with document form automatically.
1. subscriber interface module
The function of subscriber interface module can be divided into following components on realizing:
1) data show
Demonstration original analysis data and data analysis result with chart or curve mode visual pattern.
2) data management
With database mode management historical signal record.When preserving image data, preserve process conditions synchronously, information such as welding parameter, and with process conditions, information such as welding parameter deposit database in.Can conduct interviews to historical record as required, revise and backup.
2. signal and data analysis processing module
The function of signal and data analysis processing module can be divided into following components on realizing:
1) data analysis
The conventional method of analysis time series analysis; Auto-correlation and cross-correlation analysis; Fft analysis; Auto-power spectrum and crosspower spectrum analysis; Histogram amplitude probability density; Time-frequency domain analysis (short time FFT, Gabor conversion and wavelet transformation) etc.
2) sensor fusion
Adopt the linear-phase filtering technology that multiple sensor signals is handled, and by auto-correlation and cross-correlation analysis and sensor fusion techniques, the data that collect from a plurality of sensors are at first passed through preliminary treatments such as the amplification, filtering of signal conditioning circuit, carry out feature extraction then, enter the Data Fusion module at last and obtain the data fusion result.The advantage of this structure be with the Data Fusion branch for two stages---feature extraction phases and data anastomosing algorithm stage.
In feature extraction phases, we adopt traditional feature extracting method based on physical model, and the process of feature extraction depends on the result of early stage to original data signal analysis and processing.In the welding experiment in early stage, carry out statistical analysis and frequency-domain analysis (comprising short time discrete Fourier transform, Gabor spectral transformation, wavelet transformation) through the initial data of obtaining, then analyzing and the welding result of data result that data are handled and reality compares, can determine the Partial Feature of the royal purple optical signal, acoustic signals and the infrared signal that collect from sensor.Wherein apparent in view feature has the Estimation of Mean of royal purple light, infrared signal, the variance of acoustic signals is estimated, and three the road signal have characteristic frequency at the frequency band of 2～3KHz, there is characteristic frequency in acoustic signals at the frequency band of 15～20KHz, can provide 8 features to use for data anastomosing algorithm altogether.Different characteristic signal is not quite similar to the expressive ability of weld defect in different welding processes.
In the data fusion stage, we adopt the characteristic signal classification of carrying out different welding result based on the machine learning algorithm of sample, and its main method has two.First method is based on the artificial neural network of back-propagation algorithm, network structure is the structure of 8-n-1, be 8 nodes of input layer, the individual neuron of hidden layer n (4≤n≤13), 1 node of output layer can be imported each sample (8 dimension column vectors) and accurately be divided into penetration state and penetration state not.Second method is SVMs, and the kernel function of employing has linear function, polynomial function, RBF and sigmoid function.
At last, we can be at different welding processes, from 8 characteristic signals that characteristic extracting module provides, choose flexibly combination arbitrarily send into two kinds of blending algorithms in the data fusion module in any one analyze.Can search out optimum composition method by test of many times to the actual welding process at different welding material and technology.For example, in the application process of reality, if select 8 whole features, whole system can access maximum defect recognition accuracy, but feature selecting must be many more, and it is also many more that data fusion is calculated the time that is consumed.Require very high occasion in real-time, can only select the Estimation of Mean of royal purple light, infrared signal and the variance of acoustic signals to estimate this 3 features, at this moment system can reach maximum arithmetic speed, but the defect recognition accuracy also declines to a great extent thereupon.In order to obtain the balance of computational speed and recognition correct rate, can select the Estimation of Mean of royal purple light, infrared signal and acoustic signals these 5 features of frequency band, can in short operation time, obtain higher defect recognition accuracy at 2～3kHz, 15～20kHz and 30～35kHz.Improve the reliability of welding process monitoring
3) function expansion
Reserve hardware and software analysis interface, expand in order to systemic-function.
3. report generation module
The function of report generation module can be divided into following components on realizing:
1) generates analysis report automatically
Automatically analysis result (signal curve, histogram, form etc.) is embedded among WORD or the EXCEL, forms data analysis report.
2) print
Part or all of print data analysis result.
4. System self-test module
When hardware system breaks down, provide failure diagnosis information.
5. real-time monitoring modular
On the function of monitoring modular realizes in real time, can be divided into following components:
1) the unstable Realtime Alerts of welding process
2) weld defect size and position indicating
3) poor weld statistical analysis
4) typical penetration state is judged
6. network monitoring module
By cipher mode real-time image data is issued on LAN, any main frame is monitored in real time by client software in the LAN.
Program circuit is:
The flow process of the operation of entire equipment is a starting device at first, carries out the parameter setting by the user then, just can start real-time monitoring after the parameter setting is finished, the user finishes the back in monitoring Monitoring Data is saved in file, carry out the off-line data analysis then, generate form at last, whole process finishes.
Details with regard to this process describes below.
At first, starting device, at this moment equipment can start selftest module automatically and carry out the detection certainly of equipment, in case the software and hardware of equipment goes wrong, in time notifies the user to solve.
After self check is finished, enter customer parameter step is set.Equipment gets parms from database earlier and template is set carries out the selection of welding material, welding procedure and analytical parameters for the user, and the user can also make amendment to parameter and is provided with according to the operating position of reality.After the parameter setting is finished.Equipment can be automatically be saved in the parameter of this setting in the database when using for next time to be selected.
After setting up monitoring parameter, the user just can enter real-time monitoring step.Equipment at first carries out the real-time collection of data in this step, then the data that collect is carried out real-time analysis and processing.The real time data analytical procedure can be divided into again the statistical analysis of data and frequency-domain analysis, feature extraction and data fusion.Equipment at first carries out the extraction of data characteristics according to the result of statistical analysis and frequency-domain analysis, these features comprise: the Estimation of Mean of royal purple light, infrared signal, the variance of acoustic signals is estimated, three road signals are at the frequency band of 2～3KHz, and acoustic signals is at the frequency band of 15～20KHz and 30～35kHz.After having extracted feature, equipment selects the whole or a plurality of data fusion of carrying out in the above-mentioned characteristic signal to calculate according to user's setting.According to user's setting, data fusion calculating can be selected any one in following two kinds of algorithms for use: (1) based on the artificial neural network algorithm of back-propagation algorithm, network structure is the structure of 8-n-1; (2) algorithm of SVMs, the kernel function of employing has linear function, polynomial function, RBF and sigmoid function.At last, according to the data fusion result calculated, equipment generates following information: the unstable Realtime Alerts of welding process, poor weld statistical analysis, weld defect size and position indicating and typical penetration state are judged.The result that real-time data acquisition step and data analysis step obtain is sent to data disaply moudle and shows, checks for the user.
After monitoring step was finished in real time, the user can be saved in file to real-time data monitored.And the off-line data analytic function that can use equipment compares data analysis operation consuming time to data, carries out model training or the like such as Gabor analysis of spectrum, wavelet analysis and to the data blending algorithm.The result that analysis obtains according to off-line data, the user can obtain the data analysing method to specific weld material, specific weld technology optimum, at this moment can store the analytical parameters of optimum analysis method in the database into, when using same or close welding material and technology next time, can directly use.
After the off-line data analysis finishes, can generate the data analysis form, for analyzing and filing.So far, the whole monitoring flow process of equipment is promptly accused and is finished.
Claims (5)
1. high-energy beam welding process multi-signal fusion-monitoring instrument of forming by real time monitoring signals system, welding data analytical system and remote supervision system, it is characterized in that, utilize royal purple light optical pickocff, CCD optical pickocff, infrared light transducer[sensor, four groups of sensors of sonic transducer simultaneously the high energy beam current welding process to be carried out quantitative measurment; Wherein the signal of royal purple light optical pickocff, infrared light transducer[sensor and No. one sonic transducer signal are transported to the A/D capture card through signal conditioning circuit amplification and filtering respectively, the A/D capture card with the signal of the sound wave gathered, royal purple light, infrared light, is sent in the lump and is carried out data fusion in the data server simultaneously; The picture signal of ccd video camera unit is imported another data server through image pick-up card, carries out remote monitoring by LAN.
2. high-energy beam welding process multi-signal fusion-monitoring instrument according to claim 1, it is characterized in that, data fusion is divided into two stages: feature extraction phases and data are every the hop algorithm stage, the data extract stage is extracted the characteristic of the initial data that collects from sound, royal purple light, infrared light transducer[sensor, then, adopt short time discrete Fourier transform, statistic decision method, D-S evidential reasoning method and carry out data computation respectively based on the algorithm of artificial neural network.
3. high-energy beam welding process multi-signal fusion-monitoring instrument according to claim 1, it is characterized in that the condenser type sonic transducer of acoustical signal unit detects the characteristic signal of part of the audible sound below the 20KHz and the part of the ultrasonic wave more than the 20KHz simultaneously in 4～100kHz scope.
4. high-energy beam welding process multi-signal fusion-monitoring instrument according to claim 1 is characterized in that, opto-collection system adopts the off-axis acquisition mode, converges on the photodiode photosurface behind mating plate and the collector lens after filtration.
5. Data Fusion method that is used for the described high-energy beam welding process multi-signal fusion-monitoring instrument of claim 1 comprises starting device and carries out the step of equipment self-inspection; Be used for the step that the user is provided with the monitoring welding parameter; Be used for the step of monitoring in real time; The step that is used for user interface; Be used for the step that signal and data analysis are handled; The step that is used for report generation; The step that is used for network monitor; It is characterized in that the processing method of monitoring comprises in real time:
Be used for step to the welding real-time data acquisition;
Be used to comprise to the statistical analysis of data and the processing of frequency-domain analysis and data characteristics extraction and data fusion;
Carry out the extraction of data characteristics according to the result of statistical analysis and frequency-domain analysis, these features comprise: the Estimation of Mean of royal purple light, infrared signal, the variance of acoustic signals estimates that three road signals are at the frequency band of 2～3KHz, and acoustic signals is at the frequency band of 15～20KHz and 30～35kHz;
Be used for above-mentioned characteristic signal is merged the step of calculating, this step can merge calculating with a kind of in following two kinds of algorithms: (1) based on the artificial neural network algorithm of back-propagation algorithm, network structure is the structure of 8-n-1; (2) algorithm of SVMs, the kernel function of employing has linear function, polynomial function, RBF and sigmoid function;
Generate the step of following information: the unstable Realtime Alerts of welding process, poor weld statistical analysis, weld defect size and position indicating and typical penetration state are judged.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB200510080198XA CN100341658C (en) | 2005-07-04 | 2005-07-04 | High-energy beam welding process multi-signal fusion-monitoring instrument |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB200510080198XA CN100341658C (en) | 2005-07-04 | 2005-07-04 | High-energy beam welding process multi-signal fusion-monitoring instrument |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1709631A CN1709631A (en) | 2005-12-21 |
| CN100341658C true CN100341658C (en) | 2007-10-10 |
Family
ID=35705957
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB200510080198XA Expired - Fee Related CN100341658C (en) | 2005-07-04 | 2005-07-04 | High-energy beam welding process multi-signal fusion-monitoring instrument |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100341658C (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102141543A (en) * | 2010-12-28 | 2011-08-03 | 天津大学 | Method and device for detecting quality of laser welding based on microphone arrays |
Families Citing this family (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7516663B2 (en) * | 2006-11-03 | 2009-04-14 | General Electric Company | Systems and method for locating failure events in samples under load |
| DE102007024789B3 (en) * | 2007-05-26 | 2008-10-23 | Trumpf Werkzeugmaschinen Gmbh + Co. Kg | Method for detecting defects in a weld during a laser welding process |
| JP5842851B2 (en) * | 2013-03-29 | 2016-01-13 | トヨタ自動車株式会社 | Welded part inspection device and inspection method |
| CN103252573B (en) * | 2013-04-12 | 2016-01-20 | 宁波市祥和源汽配有限公司 | Coaxial vision heated filament low-power laser mould repair method and apparatus is assisted in sound emission |
| CN103464906B (en) * | 2013-10-02 | 2015-08-05 | 机械科学研究院哈尔滨焊接研究所 | Laser Welding Quality online test method |
| CN103499579B (en) * | 2013-10-02 | 2016-01-20 | 机械科学研究院哈尔滨焊接研究所 | Laser Welding Quality fast non-destructive detection method |
| CN104143183B (en) * | 2014-08-07 | 2017-12-12 | 北京理工大学 | The gray scale fusion method of visible ray and infrared black and white video image is transmitted based on brightness |
| CN104708214A (en) * | 2014-12-02 | 2015-06-17 | 苏州领创激光科技有限公司 | Laser cutting process control sampling theory and control method |
| CN105414710B (en) * | 2015-12-29 | 2017-02-01 | 哈尔滨阿尔特机器人技术有限公司 | Active and passive visual welding pool composited sensing device and sensing method realized through same |
| CN105675324B (en) * | 2016-01-20 | 2018-04-10 | 南京熊猫电子股份有限公司 | Utilize the device of ultrasound examination inverter type welder performance |
| CN105716655B (en) * | 2016-02-03 | 2018-05-01 | 西北工业大学 | Temperature and deformation real-time synchronization measuring device and method in high energy beam increasing material manufacturing |
| JP6579983B2 (en) * | 2016-03-18 | 2019-09-25 | 日立オートモティブシステムズ株式会社 | High energy beam welding quality judgment method, quality judgment device using the judgment method, and welding management system using the judgment method |
| CN106018405B (en) * | 2016-05-17 | 2019-03-22 | 机械科学研究院哈尔滨焊接研究所 | Laser welding penetration online test method |
| CN106771746B (en) * | 2016-12-20 | 2019-05-14 | 中国电器科学研究院有限公司 | A kind of electric car dynamic operation condition electromagnetic disturbance fast appraisement method |
| CN108375581B (en) * | 2017-01-04 | 2020-08-04 | 中国航空制造技术研究院 | Double-beam laser welding process defect control method based on acousto-optic signal monitoring |
| CN107414294A (en) * | 2017-08-08 | 2017-12-01 | 江苏大金激光科技有限公司 | A kind of laser-beam welding machine |
| CN108031955A (en) * | 2017-12-13 | 2018-05-15 | 太仓鼎诚电子科技有限公司 | System is monitored in a kind of welding process based on LAN |
| CN108213707B (en) * | 2018-01-26 | 2019-07-30 | 吉林大学 | Laser Welding penetration signal real-time monitoring device and method based on supersonic guide-wave |
| CN108535358A (en) * | 2018-04-10 | 2018-09-14 | 沈阳化工大学 | A kind of change wall thickness rotary work piece defect detecting device and its method |
| CN110097683A (en) * | 2018-07-20 | 2019-08-06 | 深圳怡化电脑股份有限公司 | A kind of equipment self-inspection method, apparatus, ATM and storage medium |
| CN109300116A (en) * | 2018-09-03 | 2019-02-01 | 广东工业大学 | The online defect identification method of laser welding based on machine learning |
| CN109483107A (en) * | 2018-12-29 | 2019-03-19 | 朱清 | A kind of weld seam intelligent online detection device based on Multi-source Information Fusion |
| CN110007366B (en) * | 2019-03-04 | 2020-08-25 | 中国科学院深圳先进技术研究院 | Life searching method and system based on multi-sensor fusion |
| CN110000473A (en) * | 2019-05-14 | 2019-07-12 | 广东工业大学 | A kind of galvanized steel laser adds powder welding defect online monitoring system and method |
| CN110605393B (en) * | 2019-09-25 | 2021-06-08 | 中国兵器装备集团自动化研究所 | Laser three-dimensional forming process detection method and system and application |
| CN112518122B (en) * | 2020-12-04 | 2022-05-17 | 广州德擎光学科技有限公司 | Laser processing piece fusion depth detection method, device and system |
| CN113118625A (en) * | 2021-04-23 | 2021-07-16 | 广州松兴电气股份有限公司 | Laser welding machine |
| CN113210805A (en) * | 2021-05-11 | 2021-08-06 | 浙江清华长三角研究院 | MIG welding deviation rectifying method based on industrial thermal imager and visible light camera double vision |
| CN113588074B (en) * | 2021-07-15 | 2024-04-19 | 哈尔滨工业大学（威海） | LDED on-line monitoring device based on molten pool multi-element optical information and defect diagnosis method |
| CN114913316B (en) * | 2022-04-02 | 2023-04-07 | 淮沪电力有限公司田集第二发电厂 | Image classification method and device for meter recognition of industrial equipment, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1072365A (en) * | 1991-11-20 | 1993-05-26 | 华中理工大学 | The method of real-time of welding quality in the continuous laser welding process |
| CN1154283A (en) * | 1995-10-06 | 1997-07-16 | 埃尔帕特朗尼股份公司 | Method and apparatus for monitoring and positioning bema or jet for operating workpiece |
| DE19804029A1 (en) * | 1998-02-02 | 1999-08-05 | Lesmueller Lasertechnik Gmbh | Monitoring and control device for excimer laser treatment of eyes |
| US6084203A (en) * | 1996-08-08 | 2000-07-04 | Axal | Method and device for welding with welding beam control |
| JP2001179470A (en) * | 1999-12-27 | 2001-07-03 | Sumitomo Heavy Ind Ltd | Laser beam machining state measuring apparatus |
| DE10221210A1 (en) * | 2002-05-13 | 2003-12-04 | Hydro Aluminium Deutschland | Monitor and control of laser welding has a camera as a sensor to receive the beams from the weld root, at the processing zone, to give a beam image for trouble-free identification of good/bad welds |
-
2005
- 2005-07-04 CN CNB200510080198XA patent/CN100341658C/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1072365A (en) * | 1991-11-20 | 1993-05-26 | 华中理工大学 | The method of real-time of welding quality in the continuous laser welding process |
| CN1154283A (en) * | 1995-10-06 | 1997-07-16 | 埃尔帕特朗尼股份公司 | Method and apparatus for monitoring and positioning bema or jet for operating workpiece |
| US6084203A (en) * | 1996-08-08 | 2000-07-04 | Axal | Method and device for welding with welding beam control |
| DE19804029A1 (en) * | 1998-02-02 | 1999-08-05 | Lesmueller Lasertechnik Gmbh | Monitoring and control device for excimer laser treatment of eyes |
| JP2001179470A (en) * | 1999-12-27 | 2001-07-03 | Sumitomo Heavy Ind Ltd | Laser beam machining state measuring apparatus |
| DE10221210A1 (en) * | 2002-05-13 | 2003-12-04 | Hydro Aluminium Deutschland | Monitor and control of laser welding has a camera as a sensor to receive the beams from the weld root, at the processing zone, to give a beam image for trouble-free identification of good/bad welds |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102141543A (en) * | 2010-12-28 | 2011-08-03 | 天津大学 | Method and device for detecting quality of laser welding based on microphone arrays |
| CN102141543B (en) * | 2010-12-28 | 2012-10-24 | 天津大学 | Method and device for detecting quality of laser welding based on microphone arrays |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1709631A (en) | 2005-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100341658C (en) | High-energy beam welding process multi-signal fusion-monitoring instrument | |
| CN102230952B (en) | Corona detection method based on ultraviolet photons | |
| US11564661B2 (en) | Method for optimizing ultrasonic imaging system parameter based on deep learning | |
| CN105445697B (en) | A kind of sound source direction method of low cost low-power consumption | |
| CN112285504A (en) | Multispectral solar-blind narrow-band ultraviolet imager and method for detecting different discharge states by using same | |
| RU2367549C2 (en) | Quality control method and system | |
| CN112130316B (en) | Multi-channel multi-spectral-band optical filter structure and application and method thereof | |
| CN109756720B (en) | Focus and/or parallax adjustment in acoustic imaging using distance information | |
| CN102990225A (en) | Method for detecting laser welding quality in real time | |
| CN112348052A (en) | Power transmission and transformation equipment abnormal sound source positioning method based on improved EfficientNet | |
| CN113125556A (en) | Structural damage detection system and method based on voiceprint recognition | |
| CN115186850A (en) | Dynamic monitoring method and system for submarine cable operating environment | |
| CN105891214A (en) | System And Method For Detecting A Defect In A Structure Member | |
| CN112957011A (en) | High-sensitivity weak fluorescence signal detection system, method, storage medium and application | |
| CN112308892A (en) | Shell texture analysis system and method based on LIBS technology | |
| JPH0636016A (en) | Optical inspection method and device for fault of surface of body | |
| CN114034380B (en) | One-dimensional acoustic positioning method for engine rack | |
| CN109444265B (en) | Laser ultrasonic vibration detection device and method | |
| CN214750365U (en) | Automatic calibration mechanism for fluorescence immunoassay analyzer | |
| CN210639090U (en) | Shimming Raman detection device | |
| CN113804681A (en) | Lens quality evaluation method and device based on intelligent optics | |
| CN106899919A (en) | The interception system and method for a kind of view-based access control model microphone techniques | |
| CN111007016A (en) | Device for detecting tiny particle impurities on surface of transparent material and using method | |
| CN118237778A (en) | Ultra-fast laser micro-welding quality monitoring system and monitoring method for brittle materials | |
| JP2001044532A (en) | System for monitoring quality of laser beam |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20071010 Termination date: 20140704 |
|
| EXPY | Termination of patent right or utility model |