CA2390873A1 - Method and device for quality control of the joint on sheets or strips butt-welded by means of a laser - Google Patents
Method and device for quality control of the joint on sheets or strips butt-welded by means of a laser Download PDFInfo
- Publication number
- CA2390873A1 CA2390873A1 CA002390873A CA2390873A CA2390873A1 CA 2390873 A1 CA2390873 A1 CA 2390873A1 CA 002390873 A CA002390873 A CA 002390873A CA 2390873 A CA2390873 A CA 2390873A CA 2390873 A1 CA2390873 A1 CA 2390873A1
- Authority
- CA
- Canada
- Prior art keywords
- sensor
- artificial neuronal
- network
- weld
- sensor data
- 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.)
- Abandoned
Links
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000003908 quality control method Methods 0.000 title claims abstract description 14
- 230000001537 neural effect Effects 0.000 claims abstract description 104
- 230000036961 partial effect Effects 0.000 claims abstract description 65
- 230000008569 process Effects 0.000 claims description 43
- 238000012545 processing Methods 0.000 claims description 36
- 210000002569 neuron Anatomy 0.000 claims description 30
- 230000006870 function Effects 0.000 claims description 18
- 238000012935 Averaging Methods 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 10
- 238000003860 storage Methods 0.000 claims description 10
- 230000009466 transformation Effects 0.000 claims description 10
- 230000001629 suppression Effects 0.000 claims description 5
- WQGWDDDVZFFDIG-UHFFFAOYSA-N pyrogallol Chemical compound OC1=CC=CC(O)=C1O WQGWDDDVZFFDIG-UHFFFAOYSA-N 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 abstract description 3
- 238000005259 measurement Methods 0.000 abstract 1
- 238000007781 pre-processing Methods 0.000 abstract 1
- 238000013441 quality evaluation Methods 0.000 abstract 1
- 238000003466 welding Methods 0.000 description 17
- 230000004913 activation Effects 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 4
- 230000001276 controlling effect Effects 0.000 description 4
- 238000009529 body temperature measurement Methods 0.000 description 3
- 238000001816 cooling Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 210000002364 input neuron Anatomy 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 230000001681 protective effect Effects 0.000 description 2
- 238000001303 quality assessment method Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013479 data entry Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005304 joining Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000001881 scanning electron acoustic microscopy Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- 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
- B23K26/00—Working by laser beam, e.g. welding, cutting or boring
- B23K26/02—Positioning or observing the workpiece, e.g. with respect to the point of impact; Aligning, aiming or focusing the laser beam
- B23K26/03—Observing, e.g. monitoring, the workpiece
- B23K26/032—Observing, e.g. monitoring, the workpiece using optical means
-
- 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
- B23K26/00—Working by laser beam, e.g. welding, cutting or boring
- B23K26/02—Positioning or observing the workpiece, e.g. with respect to the point of impact; Aligning, aiming or focusing the laser beam
- B23K26/03—Observing, e.g. monitoring, the workpiece
-
- 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
- B23K26/00—Working by laser beam, e.g. welding, cutting or boring
- B23K26/02—Positioning or observing the workpiece, e.g. with respect to the point of impact; Aligning, aiming or focusing the laser beam
- B23K26/03—Observing, e.g. monitoring, the workpiece
- B23K26/034—Observing the temperature of the workpiece
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/33—Director till display
- G05B2219/33025—Recurrent artificial neural network
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/33—Director till display
- G05B2219/33027—Artificial neural network controller
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/33—Director till display
- G05B2219/33038—Real time online learning, training, dynamic network
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/34—Director, elements to supervisory
- G05B2219/34153—Linear interpolation
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/37—Measurements
- G05B2219/37217—Inspect solder joint, machined part, workpiece, welding result
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/45—Nc applications
- G05B2219/45138—Laser welding
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Optics & Photonics (AREA)
- Plasma & Fusion (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- Quality & Reliability (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Laser Beam Processing (AREA)
- Lining Or Joining Of Plastics Or The Like (AREA)
Abstract
The invention relates to a method for the quality control of the joint on sheets or strips butt-welded by laser, on which a number of sensor measurements are taken, by at least two sensors arranged around the welded region. The sensor data is fed as input parameters to a combining and correlating measured-data processor for quality evaluation of the welded joint. In order to achieve a real-time quality control of the welded joint, which permits an analysis of the weld result which is true to the actual result, the stored data is fed as input parameters to at least one trainable, artificial neuronal network with an essentially hierarchical structure, which comprises at least two essentially independent artificial neuronal networks.
Furthermore, the first artificial neuronal network comprises at least two artificial neuronal networks. The first artificial neuronal network is fed the results from a data pre-processing, as input parameters and the second artificial neuronal partial network is fed the results from the first artificial neuronal partial network, as input parameters and the results from the at least one artificial neuronal network are used for quality control.
Furthermore, the first artificial neuronal network comprises at least two artificial neuronal networks. The first artificial neuronal network is fed the results from a data pre-processing, as input parameters and the second artificial neuronal partial network is fed the results from the first artificial neuronal partial network, as input parameters and the results from the at least one artificial neuronal network are used for quality control.
Description
k , m a1:
KN/so 990891US
', METHOD AND DEVICE FOR CARRYING OUT QUALITY CONTROL OF THE
SEAM ON SHEET/PLATE OR STRIP, BUTT-WELDED BY MEANS OF A
LASER DEVICE
The present invention relates to a method for quality control of the seam on sheet/plate or strip, butt-welded by means of a laser device, wherein a multitude of sensor data is measured by at least two sensors arranged around the weld location; and wherein the sensor data is fed as an input quantity to at least one summarising and correlating read-out data processing operation for assessing the quality of the weld seam; wherein the sensor data is measured by at least one sensor acquiring the weld plasma, said sensor being located at the weld location; wherein the sensor data of at least one sensor, arranged after the weld location, is acquired, said sensor acquiring the geometry of the weld seam; wherein during correlating and summarising read-out data processing, the multitude of the sensor data of the sensors, of which there are at least two, in each instance is supplied as the input quantity of at least one preliminary data processing operation; wherein the results of the preliminary data processing operation for the purpose of a co-locational relationship of the sensor data, in each instance are stored in a storage unit.
From EP 0 655 294 it is known to determine the seam quality of a laser weld seam by means of temperature measurements carried out simultaneously. To this effect, characteristic process-sensible data is determined by means of pyrometric temperature measurements, said data then being processed for quality control. In this method, temperature measurements are carried out, by means of preferably fast pyrometers, in at least two specified locations on the weld seam. The measured signals of the i individual pyrometers are correlated with each other by means of electronic signal processing. In addition, the measured values which are obtained as standard from the laser welding unit, are used and again logically linked to the measured values from the pyrometers. This summarising and correlating read-out data processing makes it possible to assess the quality of the weld seam by means of the measured process parameters.
This process is associated with the disadvantage that as the number of measured process parameters increases, due to their mutual influence on the weld process, analysis by means of a rule-based modelling is made more difficult.
From "Perspektiven der Lasermaterialbearbeitung grosser Strukturen" Sepold, Egler; in Schlusseltechnologie Laser:
Herausforderung an die Fabrik 2000; proceedings of the 12th International Congress (LASER'95); editor: Geiger M., Bamberg, Meisenbach, 1995, pp. 275-284, it is known to increase the system intelligence of laser material treatment facilities by using sensors. This enables on-line reaction to process irregularities and workpiece irregularities. To this effect, it has been proposed to arrange a sensor in front of, above, and behind the treatment location of the laser, and to feed the read-out data to a process computer.
This process is associated with the disadvantage that read-out data processing can react to known system conditions only in a static way. In the case of unknown conditions, static read-out data processing fails altogether.
i From "On-line-Qualitatskontrolle beim Metall-Schutzgasschweissen durch kiinstliche neuronale Netze";
in: Schweissen and Schneiden, 1997, vol. 2, pp. 75-80, the use of artificial neuronal networks in the field of laser weld technology is known. For quality assessment, data relating to the weld seam is processed by an artificial neuronal network. Due to their structure, neuronal networks can handle a multitude of read-out data. In addition, they can make a quality assessment of weld seams even in cases where unknown system conditions occur.
This method is associated with the disadvantage that it is not possible to assess the influence of individual read-out data on the quality of the seam. The known neuronal network only supplies go / no-go decisions as a result; the user does not know what influence the individual read-out data has on the decision finding.
It is the object of the present invention to carry out close to real-time quality control of weld seams which enables an analysis of the weld results which corresponds to the real situation.
This object is met in that the stored data is fed as input quantities to at least one trainable artificial neuronal network which comprises an essentially hierarchical network structure; that the trainable artificial neuronal network, of which there is at least one, comprises an essentially hierarchical network structure which comprises at least two essentially independent artificial neuronal partial networks; that the first artificial neuronal partial network comprises at least two independent artificial neuronal partial networks; that in each instance the results of the i preliminary data processing operations are fed as input quantities to the first artificial neuronal partial networks; that the results of the first artificial neuronal partial networks are fed as input quantities to the second artificial neuronal partial network; and that the result of the artificial neuronal network, of which there is at least one, is used for quality control.
The method according to the invention makes possible close to real-time quality control of the seam of a laser welding operation, in that it uses at least one hierarchical artificial neuronal network for processing a multitude of sensor data. It has been shown that the sensor data relating to the weld plasma and the weld seam geometry is most informative for the above purpose. Via storage units, the sensor data associated with a weld seam location, will be conveyed in parallel to the artificial neuronal network. In this way the local correlation of the signals associated with the respective weld seam locations can be carried out in close to real-time by the artificial neuronal network. By using two independent artificial neuronal networks, in the first artificial neuronal partial networks the wealth of data is reduced to a relevant minimum; this data is then correlated in the second neuronal partial network.
Furthermore, in this way it is also possible to view the results of the first artificial neuronal partial networks separate from each other and if necessary to use them for controlling various welding machine parameters.
Furthermore, these results can be stored and, should a fault in the weld seam subsequently occur, attempts can be made to find the cause of the fault by means of this data. This is advantageous in the case of product liability. By using artificial neuronal networks with a hierarchical network structure, it is possible to use a i large number of sensors for information about quality, and to evaluate the multitude of the sensor data in all their manifold combination options. Apart from the process parameters, it is also possible to evaluate machine parameters, such as parameters relating to the laser, e.g. power, mode, power distribution, focal position, as well as parameters relating to the welding facility, e.g. contact pressure force, protective gasses, seam cooling and feed rate. Furthermore, external influences which can for example affect the function of the sensors, can again flow in to form part of the assessment. This may for example relate to illumination and the temperature in the production hall. In addition, even in the case of noisy signals, the result can still be satisfactory.
Preferred is a method where in addition, a sensor for acquiring the gap geometry is used, wherein the sensor is arranged in front of the weld location. By means of this sensor, the size of the misalignment of the edges can be detected prior to welding. In this way a statement about the quality of the weld can be made independently of misalignment of the edges. In addition, by means of such information, the position of the sheets/plates in relation to each other, can be readjusted during the weld process.
A statement about the quality of the weld, depending on the gap in the joint, becomes possible by using a method in which an additional sensor arranged in front of the weld location, is used for acquiring the gap in the joint. By means of this sensor, the size of the joint gap is measured prior to welding. In addition, by means of this information, the gap in the joint can be readjusted during the welding process.
~i According to a further embodiment of the invention, a sensor is used for acquiring the weld seam temperature just behind the weld location.
Effective and fast assessment of the sensor data becomes possible if each artificial neuronal partial network comprises three layers, wherein the first layer comprises exactly one neuron, the second layer comprises a multitude of neurons, and the third layer comprises exactly one neuron. Furthermore, in this way a statement concerning the error probability, utilising a process parameter, is automated.
According to a further embodiment, the second artificial neuronal partial network comprises three layers, wherein the first layer comprises a multitude of neurons, the second layer comprises a multitude of neurons, and the third layer comprises exactly one neuron. Since there are a multitude of input neurons, the outputs of the multitude of the first neuronal partial networks can be fed in parallel to the second mutual artificial neuronal partial network. This makes possible a parallel correlation of the individual sensor data. In this way, a statement on the quality of the weld seam, taking into account a multitude of sensor data, is possible. The one output neuron of the third layer supplies a signal which makes it possible to make a statement on the quality of the seam of the laser weld.
Furthermore, it is envisaged that the learning process of the artificial neuronal network be carried out by means of a backpropagation learning algorithm in which the first artificial neuronal partial networks are adjusted at a learning rate r~ between 0.01 and 0.1 and a momentum ~i 7 _ a between 0.1 and 0.6 and in which the second neuronal partial network is adjusted at a learning rate r~ and a momentum a which are essentially adapted to a gradient of an error function of the output of the artificial neuronal partial network. The error function of the output of the artificial neuronal partial network is for example formed in that the sum of the squares of the differences between the actual output and the setpoint output is determined. During the learning phase, weighting of the individual network elements of the partial networks is adjusted such that this error function is minimal. The first and the second partial network are adapted in one operation. The network configuration which was found before training, remains unchanged even during the test phase. By means of an adaptive learning algorithm, the learning rate and the momentum of the second neuronal partial network can be adapted to the gradient of this error function which results in the global minimum of the error function being found with a high probability and with local minima of the error function being passed over.
In order to filter out, from the sensor data, information which is relevant for the statement on quality, a characteristics extraction of the sensor data is carried out during preliminary data processing. In this way, an enormous reduction in data is achieved which results in accelerated calculation in the connected systems, which makes it possible to obtain a close to real-time statement on the quality.
During characteristics extraction of the sensor data characterising the gap width, preferably the sensor data acquired by a sensor arranged in front of the weld i 8 _ location, is fed as an input quantity to an error suppression process, the results of the error suppression are fed as an input quantity to a window averaging process which essentially is freely definable, with the difference of the results of the window averaging process being established. Error suppression filters out sensor data which results due to measuring errors. Window averaging is used to suppress noise influence on the sensor data. By forming the differential value, it is possible to arrive at a statement concerning the gap width.
Furthermore, during characteristics extraction of the sensor data which characterises plasma intensity, the sensor data measured by a plasma intensity sensor can be fed as an input quantity to a window transformation. The window transformation makes it possible to filter out, from the measured plasma intensity data, the data which is relevant for assessing the seam quality.
According to a further preferred embodiment of the method according to the invention, it is envisaged that during characteristics extraction of the sensor data characterising the weld sinkage, the sensor data which was measured by a geometry sensor is fed as an input quantity to a window averaging process. By means of various window widths of window averaging, both local signal changes and signal changes indicating a trend, can be assessed in a targeted way.
A further embodiment of the method according to the invention covers a method in which for the purpose of characteristics extraction of the sensor data which reflects misalignment of edges, the sensor data measured by a geometry sensor is fed as-an input quantity to a i _ g _ mean value transformation process, and with the result of this mean value transformation process being fed to a window averaging process. Mean value transformation makes it possible to correct the signal by the entire mean value. Window averaging makes it possible to filter out changes in the misalignment of edges which changes indicate a trend. Misalignment of edges can be measured both before and after welding. In particular in the case of round blank welding it is possible, by means of only one geometry sensor, to measure the misalignment of edges before as well as after welding.
In order to limit the error value of individual sensor data to a common range of values, the results of the first artificial neuronal partial networks are standardised in relation to the range of values, and are fed to the second artificial neuronal partial network.
The maximum value of this value range can for example express a maximum error probability of the locally measured process parameter.
A further object of the invention relates to a device which is characterised in that at least two sensors for acquiring sensor data are arranged around the weld location; wherein a sensor acquiring the weld plasma is arranged at the weld location; and wherein a sensor acquiring the seam geometry is arranged after the weld location; and in that the sensor data is used as an input quantity for one preliminary data processing setup each;
in that the storage units store the results of preliminary data processing for the purpose of a co-locational relationship of the sensor data; in that the entries of the storage units are used as parallel input quantities for an essentially trainable artificial neuronal network structure; and in that the result of the ~ I
neuronal network structure is used for qualitative assessment of the weld seam. Parallel acquisition of the sensor data as well as calculation by an essentially trainable artificial neuronal network make possible close to real-time assessment of the quality of the weld seam.
In an embodiment of the device according to the invention, a geometry sensor is arranged in front of the weld location. In addition, a pyro sensor can be arranged at the weld location. The arrangement of a gap sensor in front of the weld location is also advantageous. An arrangement of several sensors makes it possible to acquire a multitude of process parameters during welding.
In addition, the quality statement depends on the number of different process parameters which are measured, wherein the device must be designed such that a multitude of sensor signals can be acquired and processed.
Below, the invention is illustrated in more detail by means of a drawing showing embodiments, as follows:
Fig. 1 a device for the quality control of a weld seam;
Fig. 2 a hierarchical network structure of an artificial neuronal network;
Fig. 3 a structure of the first artificial neuronal partial network;
Fig. 4 a window transformation of the sensor data;
Fig. 5 a structure of the second artificial neuronal partial network;
Fig. 6 a diagrammatic representation of a method according to the invention.
Figure 1 shows a first embodiment of a device for carrying out the quality control of a seam on sheet/plate or strip, butt-welded by means of a laser device. Two pieces of sheet/plate or strip 100, 102, to be butt-welded, are transported in the direction of conveyance F
in a transporting and joining device (not shown) at a specified gap 104 in the joint, underneath a weld head 112 of a laser welding plant. In the region of the weld head 112, the two pieces of sheet/plate 100, 102 are butt-welded together in a weld seam 106 by means of a laser beam L.
Along the weld seam 106 or the gap 104 in the joint, sensors 108, 110, 114 and 116 are arranged. Sensor 108 acquires the geometry of the gap 104 in the joint in front of the weld. During this process, the vertical misalignment of edges of the sheet/plate is measured by sensor 108. Sensor 110 acquires the gap width of the gap 104 in the joint. During this process, the space between the pieces of sheet/plate 100, 102 is measured by sensors suitable for this purpose which operate for example according to the light-section process or the transmitted light process. The sensor 114 is used to acquire the plasma intensity of the laser weld beam L. The geometry sensor 116 is used to acquire the misalignment of the edge as well as the weld sinkage of the weld seam 106 behind the weld. In the case of round blank welding (not shown), the geometry of the gap 104 in the joint and of the weld seam 106 can only be acquired by a sensor.
Moreover, the weld temperature can be determined by means of a pyro sensor (not shown).
i The sensor data acquired by the sensors 108, 110, 114 and 116, is regularly polled by preliminary data processing units 118, 120, 122 and 124. The polling frequency of the individual preliminary data processing units is between a few hertz and a few kilohertz.
The data measured by sensor 108 is read in by the preliminary data processing unit 118 at regular intervals. As far as the misalignment of edges which is measured by sensor 108 is concerned, by means of window averaging, the arithmetic mean of this sensor data across the window width is calculated. The calculated arithmetic mean values of the individual windows are corrected by the amount of the total mean value.
The sensor data read in by the preliminary data processing unit 120 of the gap sensor 110 for the position of the right hand as well as the left hand edge of the gap 104 in the joint, is reconstructed by means of interpolation and linear polynomials because, due to incorrect measuring, the sensor signals show considerable deviations from the actual course of the seam. By means of window averaging, an arithmetic mean value is formed from the reconstructed sensor data. From the arithmetic mean values, obtained in this way, of the sensor data from the right as well as the left edge of the gap 104 in the joint, the difference is calculated which provides information about the size of gap widening.
The sensor data measured by the plasma sensor 114 is read in by the preliminary data processing unit 122, said data being processed as shown in Figure 4. During this process, the sensor data is fed to the input 400. From this sensor data, by means of window averaging, the arithmetic mean value of the respective last 10 sensor i1 data entries measured, is calculated in unit 402. In unit 404, the difference between the actually measured sensor value and the result of unit 402 is calculated. In unit 406 the entire mean value of the output signal of unit 404 is calculated. From the output value of unit 406, the global standard deviation is calculated in unit 408. In unit 410,the arithmetic mean value of the last 10 results of unit 404 is calculated by means of window averaging.
By means of the results of units 410 and 404, unit 412 calculates the local standard deviation. In unit 419, the maximum difference between the output of unit 404 and the output of unit 410 is calculated. In unit 416 the result of preliminary data processing is calculated such that the result of unit 414 is multiplied by the result of unit 412, with the resulting value being divided by the result of unit 408.
The sensor data of sensor 116, said data having been read into the preliminary data processing unit 124, is subjected to a window transformation process for mean value calculation. In this way, window width can be set such that local changes as well as changes indicating a trend, of the weld sinkage of the weld seam 106, make themselves felt in the mean value calculation. In the case of a temporary change in the weld sinkage, at a window width of ten data points, there is a more pronounced change in the arithmetic mean value than at a window width of forty data points.
The values calculated by means of the preliminary data processing units 118, 120, 122 and 124 are stored in storage units 119, 121, 123 and 125. By means of these storage units it is possible to feed the data from all sensors belonging to one and the same weld seam point, concurrently to the artificial neuronal network 128.
Since the sensor data of a weld seam point is measured by sensors 108 to 116 at different points in time, the sensor data is polled at different intervals by the preliminary data processing units 118, 120, 122 and 124, and preliminary data processing of the various sensor data requires a different effort, the data concerning a weld seam point is not present concurrently at the output of the preliminary data processing units 118, 120, 122 and 124. By storing the data in units 119, 121, 123 and 125, this offset in time is compensated for, so that the artificial neuronal network 128 receives data about a weld seam point also at one point in time. Only in this way is correlation of the sensor data by the artificial neuronal network possible. The storage units 119, 121, 123 and 125 are controlled by a common clock signal CLK.
The stored data belonging to one weld seam point, is fed to the trainable artificial neuronal network 128 as soon as the clock signal is present. In this artificial neuronal network 128, an output signal 130 is calculated from the data of the preliminary data processing units 118, 120, 122 and 124, with a statement about the quality of the weld seam being made by means of said output signal 130.
Figure 2 shows the trainable artificial neuronal network 128 with an essentially hierarchical network structure.
The neuronal network 128 comprises a multitude of first artificial neuronal partial networks 21$, 220, 222 and 224 as well as a second artificial neuronal partial network 242. The results of the preliminary data processing units 118, 120, 122 and 124 are fed to a first artificial neuronal partial network 218, 220, 222 or 224.
The network structure of the first artificial neuronal partial networks 218, 220, 224 and 226 is shown in Figure i r 3. It comprises an input layer 316, a covered layer 318 as well as an output layer 320. The input layer 316 comprises an input neuron 300. The input value 301 of the first artificial neuronal partial network, said input value being an output value of a preliminary data processing operation, is distributed by the input neuron 300, to the neurons of the covered layer 318 with different weighting. The covered layer 318 comprises a multitude of neurons 302 - 312.
By means of an activation function, the value of the output signal is determined from the weighted input signals of each individual neuron. The activation function can for example be a sigmoidal function or a tangent hyperbolic function.
From the sum of the weighted input signals and a threshold value as an input value of an activation function, the value of the output signal is calculated as follows:
Yj = Fj (~ C~i~ * Xi +
where:
Y~ - value of the output signal;
F~ - activation function;
c~i~ - weighting of the input signals;
Xi - input signal of the neuron; and - threshold value of the neuron.
The output signals of the neurons of the covered layer 318 are fed to the neuron of the output layer 314. Here too, by means of weighting of the input signals, an I
activation function and a threshold value, the value of the output signal 316 is calculated.
The outputs of the first artificial neuronal partial networks 218 - 224 are standardised in relation to the range of values. By such standardisation in relation to the range of values, the outputs of the first artificial neuronal partial networks 218 - 224 are standardised to a range of values of e.g. 0 - 1. The outputs of the first artificial neuronal partial networks 218 - 224 can thus be considered to be local error values. Thus, for example an output value of 1 of the artificial neuronal partial network 224, means that the error probability in the case of weld sinkage is 100 0. An output value of 0.5 of the artificial neuronal partial network 218 means that the error probability of the process parameter 'misalignment of edges', allocated to the first artificial neuronal partial network 218, is 50 ~, said process parameter having been measured by a geometry sensor.
The individual local error probabilities are fed in parallel to the second artificial neuronal partial network 242.
The structure of the second artificial neuronal partial network 242 is shown in Figure 5. The second artificial neuronal partial network 242 comprises an input layer 516, a covered layer 518 as well as an output layer 520.
The input layer 516 comprises a multitude of neurons 532, 534, 536 and 538. The respective output values of the standardisation in relation to the range of values of the first artificial neuronal partial networks 232 - 238, are used as the input values of the neurons of the input layer 516 of the second artificial neuronal partial network 242. The value of the output signals is i calculated in the neurons 532 - 538, again by means of an activation function as well as weighting of the input signal and a limiting value, as has been described above by means of Fig. 3.
In the covered layer 518 there are a multitude of neurons 502 - 512. The output signals of all neurons 532 - 538 of the input layer 516 are fed to said neurons 502 - 512.
The inputs of all neurons 502 to 538 are weighted. The output signals of neurons 502 to 538 are calculated by means of an activation function and a limiting value.
In the output layer 520 there is only one neuron 514, to which the output signals of neurons 502 - 512 of the covered layer 518 are fed. Here again, the input signals are weighted, the sum of the weighted input signals is added to a threshold value, and the result is used as an input value for an activation function. The resulting output signal is used for assessing the quality of the weld seam. During a training phase, individual weighting of the input signals of the neurons in the first artificial neuronal partial networks 218 - 224 as well as the second artificial neuronal partial networks 242 are arranged such that the statement of the artificial neuronal network 128 simulates the statement of a manual observer. By means of reference welding, a setpoint value / actual value comparison can be carried out at the output of the artificial neuronal network 128, with weighting occurring by means of the backpropagation learning algorithm.
From the multitude of input values, the second artificial neuronal partial network 242 determines an output value 244 which makes it possible to make a statement about the i weld seam quality. The reciprocal action between the individual process parameters is taken into account by the second artificial neuronal partial network 242. It is quite possible that the error probability of the process parameter 'weld sinkage' is around 80 %, but because of the interaction with other process parameters the total error probability of the weld seam is around 10 0.
Fig. 6 shows the individual process steps in their logical sequence. First of all sensor data 600 is acquired by various sensors such as e.g. geometry sensors, gap-width sensors, gyro sensors as well as plasma sensors. This sensor data is conveyed for data analysis by the artificial neuronal networks 602. By means of the values determined by the artificial neuronal networks, an assessment of the seam 604 becomes possible.
Furthermore it is possible to make a statement about the condition of the welding facility, e.g. feed rate, seam cooling, weld power or contact pressure force. The results from assessment of the seam as well as the sensor data are stored in a database 608. The data records stored in the database 608 are used for product evaluation and for plant evaluation. Furthermore, this data is used for proof, should any subsequent product liability issues arise. In addition, this data can be used as proof for quality certification.
Furthermore, the results of seam assessment 604 are used to build a closed-loop control circuit 610. The data is used for controlling plant engineering 603a as well as for controlling the laser technology 603b. Plant engineering comprises settings of the welding facility such as contact pressure force, the supply of protective gasses and subsequent cooling as well as the feed rate of the pieces of sheet/plate to be welded together.
i _ 19 _ Controlling the laser technology 603b comprises control of the weld power, the weld temperature, the power distribution as well as the focal position of the welding beam. By means of the method according to the invention it is thus possible not only to make statements about the product quality but also to set on-line plant parameters.
KN/so 990891US
', METHOD AND DEVICE FOR CARRYING OUT QUALITY CONTROL OF THE
SEAM ON SHEET/PLATE OR STRIP, BUTT-WELDED BY MEANS OF A
LASER DEVICE
The present invention relates to a method for quality control of the seam on sheet/plate or strip, butt-welded by means of a laser device, wherein a multitude of sensor data is measured by at least two sensors arranged around the weld location; and wherein the sensor data is fed as an input quantity to at least one summarising and correlating read-out data processing operation for assessing the quality of the weld seam; wherein the sensor data is measured by at least one sensor acquiring the weld plasma, said sensor being located at the weld location; wherein the sensor data of at least one sensor, arranged after the weld location, is acquired, said sensor acquiring the geometry of the weld seam; wherein during correlating and summarising read-out data processing, the multitude of the sensor data of the sensors, of which there are at least two, in each instance is supplied as the input quantity of at least one preliminary data processing operation; wherein the results of the preliminary data processing operation for the purpose of a co-locational relationship of the sensor data, in each instance are stored in a storage unit.
From EP 0 655 294 it is known to determine the seam quality of a laser weld seam by means of temperature measurements carried out simultaneously. To this effect, characteristic process-sensible data is determined by means of pyrometric temperature measurements, said data then being processed for quality control. In this method, temperature measurements are carried out, by means of preferably fast pyrometers, in at least two specified locations on the weld seam. The measured signals of the i individual pyrometers are correlated with each other by means of electronic signal processing. In addition, the measured values which are obtained as standard from the laser welding unit, are used and again logically linked to the measured values from the pyrometers. This summarising and correlating read-out data processing makes it possible to assess the quality of the weld seam by means of the measured process parameters.
This process is associated with the disadvantage that as the number of measured process parameters increases, due to their mutual influence on the weld process, analysis by means of a rule-based modelling is made more difficult.
From "Perspektiven der Lasermaterialbearbeitung grosser Strukturen" Sepold, Egler; in Schlusseltechnologie Laser:
Herausforderung an die Fabrik 2000; proceedings of the 12th International Congress (LASER'95); editor: Geiger M., Bamberg, Meisenbach, 1995, pp. 275-284, it is known to increase the system intelligence of laser material treatment facilities by using sensors. This enables on-line reaction to process irregularities and workpiece irregularities. To this effect, it has been proposed to arrange a sensor in front of, above, and behind the treatment location of the laser, and to feed the read-out data to a process computer.
This process is associated with the disadvantage that read-out data processing can react to known system conditions only in a static way. In the case of unknown conditions, static read-out data processing fails altogether.
i From "On-line-Qualitatskontrolle beim Metall-Schutzgasschweissen durch kiinstliche neuronale Netze";
in: Schweissen and Schneiden, 1997, vol. 2, pp. 75-80, the use of artificial neuronal networks in the field of laser weld technology is known. For quality assessment, data relating to the weld seam is processed by an artificial neuronal network. Due to their structure, neuronal networks can handle a multitude of read-out data. In addition, they can make a quality assessment of weld seams even in cases where unknown system conditions occur.
This method is associated with the disadvantage that it is not possible to assess the influence of individual read-out data on the quality of the seam. The known neuronal network only supplies go / no-go decisions as a result; the user does not know what influence the individual read-out data has on the decision finding.
It is the object of the present invention to carry out close to real-time quality control of weld seams which enables an analysis of the weld results which corresponds to the real situation.
This object is met in that the stored data is fed as input quantities to at least one trainable artificial neuronal network which comprises an essentially hierarchical network structure; that the trainable artificial neuronal network, of which there is at least one, comprises an essentially hierarchical network structure which comprises at least two essentially independent artificial neuronal partial networks; that the first artificial neuronal partial network comprises at least two independent artificial neuronal partial networks; that in each instance the results of the i preliminary data processing operations are fed as input quantities to the first artificial neuronal partial networks; that the results of the first artificial neuronal partial networks are fed as input quantities to the second artificial neuronal partial network; and that the result of the artificial neuronal network, of which there is at least one, is used for quality control.
The method according to the invention makes possible close to real-time quality control of the seam of a laser welding operation, in that it uses at least one hierarchical artificial neuronal network for processing a multitude of sensor data. It has been shown that the sensor data relating to the weld plasma and the weld seam geometry is most informative for the above purpose. Via storage units, the sensor data associated with a weld seam location, will be conveyed in parallel to the artificial neuronal network. In this way the local correlation of the signals associated with the respective weld seam locations can be carried out in close to real-time by the artificial neuronal network. By using two independent artificial neuronal networks, in the first artificial neuronal partial networks the wealth of data is reduced to a relevant minimum; this data is then correlated in the second neuronal partial network.
Furthermore, in this way it is also possible to view the results of the first artificial neuronal partial networks separate from each other and if necessary to use them for controlling various welding machine parameters.
Furthermore, these results can be stored and, should a fault in the weld seam subsequently occur, attempts can be made to find the cause of the fault by means of this data. This is advantageous in the case of product liability. By using artificial neuronal networks with a hierarchical network structure, it is possible to use a i large number of sensors for information about quality, and to evaluate the multitude of the sensor data in all their manifold combination options. Apart from the process parameters, it is also possible to evaluate machine parameters, such as parameters relating to the laser, e.g. power, mode, power distribution, focal position, as well as parameters relating to the welding facility, e.g. contact pressure force, protective gasses, seam cooling and feed rate. Furthermore, external influences which can for example affect the function of the sensors, can again flow in to form part of the assessment. This may for example relate to illumination and the temperature in the production hall. In addition, even in the case of noisy signals, the result can still be satisfactory.
Preferred is a method where in addition, a sensor for acquiring the gap geometry is used, wherein the sensor is arranged in front of the weld location. By means of this sensor, the size of the misalignment of the edges can be detected prior to welding. In this way a statement about the quality of the weld can be made independently of misalignment of the edges. In addition, by means of such information, the position of the sheets/plates in relation to each other, can be readjusted during the weld process.
A statement about the quality of the weld, depending on the gap in the joint, becomes possible by using a method in which an additional sensor arranged in front of the weld location, is used for acquiring the gap in the joint. By means of this sensor, the size of the joint gap is measured prior to welding. In addition, by means of this information, the gap in the joint can be readjusted during the welding process.
~i According to a further embodiment of the invention, a sensor is used for acquiring the weld seam temperature just behind the weld location.
Effective and fast assessment of the sensor data becomes possible if each artificial neuronal partial network comprises three layers, wherein the first layer comprises exactly one neuron, the second layer comprises a multitude of neurons, and the third layer comprises exactly one neuron. Furthermore, in this way a statement concerning the error probability, utilising a process parameter, is automated.
According to a further embodiment, the second artificial neuronal partial network comprises three layers, wherein the first layer comprises a multitude of neurons, the second layer comprises a multitude of neurons, and the third layer comprises exactly one neuron. Since there are a multitude of input neurons, the outputs of the multitude of the first neuronal partial networks can be fed in parallel to the second mutual artificial neuronal partial network. This makes possible a parallel correlation of the individual sensor data. In this way, a statement on the quality of the weld seam, taking into account a multitude of sensor data, is possible. The one output neuron of the third layer supplies a signal which makes it possible to make a statement on the quality of the seam of the laser weld.
Furthermore, it is envisaged that the learning process of the artificial neuronal network be carried out by means of a backpropagation learning algorithm in which the first artificial neuronal partial networks are adjusted at a learning rate r~ between 0.01 and 0.1 and a momentum ~i 7 _ a between 0.1 and 0.6 and in which the second neuronal partial network is adjusted at a learning rate r~ and a momentum a which are essentially adapted to a gradient of an error function of the output of the artificial neuronal partial network. The error function of the output of the artificial neuronal partial network is for example formed in that the sum of the squares of the differences between the actual output and the setpoint output is determined. During the learning phase, weighting of the individual network elements of the partial networks is adjusted such that this error function is minimal. The first and the second partial network are adapted in one operation. The network configuration which was found before training, remains unchanged even during the test phase. By means of an adaptive learning algorithm, the learning rate and the momentum of the second neuronal partial network can be adapted to the gradient of this error function which results in the global minimum of the error function being found with a high probability and with local minima of the error function being passed over.
In order to filter out, from the sensor data, information which is relevant for the statement on quality, a characteristics extraction of the sensor data is carried out during preliminary data processing. In this way, an enormous reduction in data is achieved which results in accelerated calculation in the connected systems, which makes it possible to obtain a close to real-time statement on the quality.
During characteristics extraction of the sensor data characterising the gap width, preferably the sensor data acquired by a sensor arranged in front of the weld i 8 _ location, is fed as an input quantity to an error suppression process, the results of the error suppression are fed as an input quantity to a window averaging process which essentially is freely definable, with the difference of the results of the window averaging process being established. Error suppression filters out sensor data which results due to measuring errors. Window averaging is used to suppress noise influence on the sensor data. By forming the differential value, it is possible to arrive at a statement concerning the gap width.
Furthermore, during characteristics extraction of the sensor data which characterises plasma intensity, the sensor data measured by a plasma intensity sensor can be fed as an input quantity to a window transformation. The window transformation makes it possible to filter out, from the measured plasma intensity data, the data which is relevant for assessing the seam quality.
According to a further preferred embodiment of the method according to the invention, it is envisaged that during characteristics extraction of the sensor data characterising the weld sinkage, the sensor data which was measured by a geometry sensor is fed as an input quantity to a window averaging process. By means of various window widths of window averaging, both local signal changes and signal changes indicating a trend, can be assessed in a targeted way.
A further embodiment of the method according to the invention covers a method in which for the purpose of characteristics extraction of the sensor data which reflects misalignment of edges, the sensor data measured by a geometry sensor is fed as-an input quantity to a i _ g _ mean value transformation process, and with the result of this mean value transformation process being fed to a window averaging process. Mean value transformation makes it possible to correct the signal by the entire mean value. Window averaging makes it possible to filter out changes in the misalignment of edges which changes indicate a trend. Misalignment of edges can be measured both before and after welding. In particular in the case of round blank welding it is possible, by means of only one geometry sensor, to measure the misalignment of edges before as well as after welding.
In order to limit the error value of individual sensor data to a common range of values, the results of the first artificial neuronal partial networks are standardised in relation to the range of values, and are fed to the second artificial neuronal partial network.
The maximum value of this value range can for example express a maximum error probability of the locally measured process parameter.
A further object of the invention relates to a device which is characterised in that at least two sensors for acquiring sensor data are arranged around the weld location; wherein a sensor acquiring the weld plasma is arranged at the weld location; and wherein a sensor acquiring the seam geometry is arranged after the weld location; and in that the sensor data is used as an input quantity for one preliminary data processing setup each;
in that the storage units store the results of preliminary data processing for the purpose of a co-locational relationship of the sensor data; in that the entries of the storage units are used as parallel input quantities for an essentially trainable artificial neuronal network structure; and in that the result of the ~ I
neuronal network structure is used for qualitative assessment of the weld seam. Parallel acquisition of the sensor data as well as calculation by an essentially trainable artificial neuronal network make possible close to real-time assessment of the quality of the weld seam.
In an embodiment of the device according to the invention, a geometry sensor is arranged in front of the weld location. In addition, a pyro sensor can be arranged at the weld location. The arrangement of a gap sensor in front of the weld location is also advantageous. An arrangement of several sensors makes it possible to acquire a multitude of process parameters during welding.
In addition, the quality statement depends on the number of different process parameters which are measured, wherein the device must be designed such that a multitude of sensor signals can be acquired and processed.
Below, the invention is illustrated in more detail by means of a drawing showing embodiments, as follows:
Fig. 1 a device for the quality control of a weld seam;
Fig. 2 a hierarchical network structure of an artificial neuronal network;
Fig. 3 a structure of the first artificial neuronal partial network;
Fig. 4 a window transformation of the sensor data;
Fig. 5 a structure of the second artificial neuronal partial network;
Fig. 6 a diagrammatic representation of a method according to the invention.
Figure 1 shows a first embodiment of a device for carrying out the quality control of a seam on sheet/plate or strip, butt-welded by means of a laser device. Two pieces of sheet/plate or strip 100, 102, to be butt-welded, are transported in the direction of conveyance F
in a transporting and joining device (not shown) at a specified gap 104 in the joint, underneath a weld head 112 of a laser welding plant. In the region of the weld head 112, the two pieces of sheet/plate 100, 102 are butt-welded together in a weld seam 106 by means of a laser beam L.
Along the weld seam 106 or the gap 104 in the joint, sensors 108, 110, 114 and 116 are arranged. Sensor 108 acquires the geometry of the gap 104 in the joint in front of the weld. During this process, the vertical misalignment of edges of the sheet/plate is measured by sensor 108. Sensor 110 acquires the gap width of the gap 104 in the joint. During this process, the space between the pieces of sheet/plate 100, 102 is measured by sensors suitable for this purpose which operate for example according to the light-section process or the transmitted light process. The sensor 114 is used to acquire the plasma intensity of the laser weld beam L. The geometry sensor 116 is used to acquire the misalignment of the edge as well as the weld sinkage of the weld seam 106 behind the weld. In the case of round blank welding (not shown), the geometry of the gap 104 in the joint and of the weld seam 106 can only be acquired by a sensor.
Moreover, the weld temperature can be determined by means of a pyro sensor (not shown).
i The sensor data acquired by the sensors 108, 110, 114 and 116, is regularly polled by preliminary data processing units 118, 120, 122 and 124. The polling frequency of the individual preliminary data processing units is between a few hertz and a few kilohertz.
The data measured by sensor 108 is read in by the preliminary data processing unit 118 at regular intervals. As far as the misalignment of edges which is measured by sensor 108 is concerned, by means of window averaging, the arithmetic mean of this sensor data across the window width is calculated. The calculated arithmetic mean values of the individual windows are corrected by the amount of the total mean value.
The sensor data read in by the preliminary data processing unit 120 of the gap sensor 110 for the position of the right hand as well as the left hand edge of the gap 104 in the joint, is reconstructed by means of interpolation and linear polynomials because, due to incorrect measuring, the sensor signals show considerable deviations from the actual course of the seam. By means of window averaging, an arithmetic mean value is formed from the reconstructed sensor data. From the arithmetic mean values, obtained in this way, of the sensor data from the right as well as the left edge of the gap 104 in the joint, the difference is calculated which provides information about the size of gap widening.
The sensor data measured by the plasma sensor 114 is read in by the preliminary data processing unit 122, said data being processed as shown in Figure 4. During this process, the sensor data is fed to the input 400. From this sensor data, by means of window averaging, the arithmetic mean value of the respective last 10 sensor i1 data entries measured, is calculated in unit 402. In unit 404, the difference between the actually measured sensor value and the result of unit 402 is calculated. In unit 406 the entire mean value of the output signal of unit 404 is calculated. From the output value of unit 406, the global standard deviation is calculated in unit 408. In unit 410,the arithmetic mean value of the last 10 results of unit 404 is calculated by means of window averaging.
By means of the results of units 410 and 404, unit 412 calculates the local standard deviation. In unit 419, the maximum difference between the output of unit 404 and the output of unit 410 is calculated. In unit 416 the result of preliminary data processing is calculated such that the result of unit 414 is multiplied by the result of unit 412, with the resulting value being divided by the result of unit 408.
The sensor data of sensor 116, said data having been read into the preliminary data processing unit 124, is subjected to a window transformation process for mean value calculation. In this way, window width can be set such that local changes as well as changes indicating a trend, of the weld sinkage of the weld seam 106, make themselves felt in the mean value calculation. In the case of a temporary change in the weld sinkage, at a window width of ten data points, there is a more pronounced change in the arithmetic mean value than at a window width of forty data points.
The values calculated by means of the preliminary data processing units 118, 120, 122 and 124 are stored in storage units 119, 121, 123 and 125. By means of these storage units it is possible to feed the data from all sensors belonging to one and the same weld seam point, concurrently to the artificial neuronal network 128.
Since the sensor data of a weld seam point is measured by sensors 108 to 116 at different points in time, the sensor data is polled at different intervals by the preliminary data processing units 118, 120, 122 and 124, and preliminary data processing of the various sensor data requires a different effort, the data concerning a weld seam point is not present concurrently at the output of the preliminary data processing units 118, 120, 122 and 124. By storing the data in units 119, 121, 123 and 125, this offset in time is compensated for, so that the artificial neuronal network 128 receives data about a weld seam point also at one point in time. Only in this way is correlation of the sensor data by the artificial neuronal network possible. The storage units 119, 121, 123 and 125 are controlled by a common clock signal CLK.
The stored data belonging to one weld seam point, is fed to the trainable artificial neuronal network 128 as soon as the clock signal is present. In this artificial neuronal network 128, an output signal 130 is calculated from the data of the preliminary data processing units 118, 120, 122 and 124, with a statement about the quality of the weld seam being made by means of said output signal 130.
Figure 2 shows the trainable artificial neuronal network 128 with an essentially hierarchical network structure.
The neuronal network 128 comprises a multitude of first artificial neuronal partial networks 21$, 220, 222 and 224 as well as a second artificial neuronal partial network 242. The results of the preliminary data processing units 118, 120, 122 and 124 are fed to a first artificial neuronal partial network 218, 220, 222 or 224.
The network structure of the first artificial neuronal partial networks 218, 220, 224 and 226 is shown in Figure i r 3. It comprises an input layer 316, a covered layer 318 as well as an output layer 320. The input layer 316 comprises an input neuron 300. The input value 301 of the first artificial neuronal partial network, said input value being an output value of a preliminary data processing operation, is distributed by the input neuron 300, to the neurons of the covered layer 318 with different weighting. The covered layer 318 comprises a multitude of neurons 302 - 312.
By means of an activation function, the value of the output signal is determined from the weighted input signals of each individual neuron. The activation function can for example be a sigmoidal function or a tangent hyperbolic function.
From the sum of the weighted input signals and a threshold value as an input value of an activation function, the value of the output signal is calculated as follows:
Yj = Fj (~ C~i~ * Xi +
where:
Y~ - value of the output signal;
F~ - activation function;
c~i~ - weighting of the input signals;
Xi - input signal of the neuron; and - threshold value of the neuron.
The output signals of the neurons of the covered layer 318 are fed to the neuron of the output layer 314. Here too, by means of weighting of the input signals, an I
activation function and a threshold value, the value of the output signal 316 is calculated.
The outputs of the first artificial neuronal partial networks 218 - 224 are standardised in relation to the range of values. By such standardisation in relation to the range of values, the outputs of the first artificial neuronal partial networks 218 - 224 are standardised to a range of values of e.g. 0 - 1. The outputs of the first artificial neuronal partial networks 218 - 224 can thus be considered to be local error values. Thus, for example an output value of 1 of the artificial neuronal partial network 224, means that the error probability in the case of weld sinkage is 100 0. An output value of 0.5 of the artificial neuronal partial network 218 means that the error probability of the process parameter 'misalignment of edges', allocated to the first artificial neuronal partial network 218, is 50 ~, said process parameter having been measured by a geometry sensor.
The individual local error probabilities are fed in parallel to the second artificial neuronal partial network 242.
The structure of the second artificial neuronal partial network 242 is shown in Figure 5. The second artificial neuronal partial network 242 comprises an input layer 516, a covered layer 518 as well as an output layer 520.
The input layer 516 comprises a multitude of neurons 532, 534, 536 and 538. The respective output values of the standardisation in relation to the range of values of the first artificial neuronal partial networks 232 - 238, are used as the input values of the neurons of the input layer 516 of the second artificial neuronal partial network 242. The value of the output signals is i calculated in the neurons 532 - 538, again by means of an activation function as well as weighting of the input signal and a limiting value, as has been described above by means of Fig. 3.
In the covered layer 518 there are a multitude of neurons 502 - 512. The output signals of all neurons 532 - 538 of the input layer 516 are fed to said neurons 502 - 512.
The inputs of all neurons 502 to 538 are weighted. The output signals of neurons 502 to 538 are calculated by means of an activation function and a limiting value.
In the output layer 520 there is only one neuron 514, to which the output signals of neurons 502 - 512 of the covered layer 518 are fed. Here again, the input signals are weighted, the sum of the weighted input signals is added to a threshold value, and the result is used as an input value for an activation function. The resulting output signal is used for assessing the quality of the weld seam. During a training phase, individual weighting of the input signals of the neurons in the first artificial neuronal partial networks 218 - 224 as well as the second artificial neuronal partial networks 242 are arranged such that the statement of the artificial neuronal network 128 simulates the statement of a manual observer. By means of reference welding, a setpoint value / actual value comparison can be carried out at the output of the artificial neuronal network 128, with weighting occurring by means of the backpropagation learning algorithm.
From the multitude of input values, the second artificial neuronal partial network 242 determines an output value 244 which makes it possible to make a statement about the i weld seam quality. The reciprocal action between the individual process parameters is taken into account by the second artificial neuronal partial network 242. It is quite possible that the error probability of the process parameter 'weld sinkage' is around 80 %, but because of the interaction with other process parameters the total error probability of the weld seam is around 10 0.
Fig. 6 shows the individual process steps in their logical sequence. First of all sensor data 600 is acquired by various sensors such as e.g. geometry sensors, gap-width sensors, gyro sensors as well as plasma sensors. This sensor data is conveyed for data analysis by the artificial neuronal networks 602. By means of the values determined by the artificial neuronal networks, an assessment of the seam 604 becomes possible.
Furthermore it is possible to make a statement about the condition of the welding facility, e.g. feed rate, seam cooling, weld power or contact pressure force. The results from assessment of the seam as well as the sensor data are stored in a database 608. The data records stored in the database 608 are used for product evaluation and for plant evaluation. Furthermore, this data is used for proof, should any subsequent product liability issues arise. In addition, this data can be used as proof for quality certification.
Furthermore, the results of seam assessment 604 are used to build a closed-loop control circuit 610. The data is used for controlling plant engineering 603a as well as for controlling the laser technology 603b. Plant engineering comprises settings of the welding facility such as contact pressure force, the supply of protective gasses and subsequent cooling as well as the feed rate of the pieces of sheet/plate to be welded together.
i _ 19 _ Controlling the laser technology 603b comprises control of the weld power, the weld temperature, the power distribution as well as the focal position of the welding beam. By means of the method according to the invention it is thus possible not only to make statements about the product quality but also to set on-line plant parameters.
Claims (17)
1. A method for quality control of the seam on sheet/plate or strip, butt-welded by means of a laser device - wherein a multitude of sensor data is measured by at least two sensors arranged around t:he weld location;
- wherein the sensor data is fed as an input quantity to at least one summarising and correlating read-out data processing operation for assessing the quality of the weld seam;
- wherein the sensor data is measured by at least one sensor acquiring the weld plasma, said sensor being located at the weld location;
- wherein the sensor data of at least one sensor, arranged after the weld location, is acquired, said sensor acquiring the geometry of the weld seam;
- wherein during correlating and summarising read-out data processing, the multitude of the sensor data of the sensors, of which there are at least two, in each instance is supplied as the input quantity of at least one preliminary data processing operation;
- wherein the results of the preliminary data processing operation for the purpose of a co-locational relationship of the sensor data, in each instance are stored in a storage unit;
characterised in that - the stored data is fed as input quantities to at least one trainable artificial neuronal network which comprises an essentially hierarchical network structure;
- the trainable artificial neuronal network, of which there is at least one, comprises an essentially hierarchical network structure which comprises at least two essentially independent, trainable artificial neuronal partial networks;
- the first artificial neuronal partial network comprises at least two independent artificial neuronal partial networks;
in each instance the results of the preliminary data processing operations are fed as input quantities to the first artificial neuronal partial networks;
- the results of the first artificial neuronal partial networks are fed as input quantities to the second artificial neuronal partial network;
and - the result of the artificial neuronal network, of which there is at least one, is used for quality control.
- wherein the sensor data is fed as an input quantity to at least one summarising and correlating read-out data processing operation for assessing the quality of the weld seam;
- wherein the sensor data is measured by at least one sensor acquiring the weld plasma, said sensor being located at the weld location;
- wherein the sensor data of at least one sensor, arranged after the weld location, is acquired, said sensor acquiring the geometry of the weld seam;
- wherein during correlating and summarising read-out data processing, the multitude of the sensor data of the sensors, of which there are at least two, in each instance is supplied as the input quantity of at least one preliminary data processing operation;
- wherein the results of the preliminary data processing operation for the purpose of a co-locational relationship of the sensor data, in each instance are stored in a storage unit;
characterised in that - the stored data is fed as input quantities to at least one trainable artificial neuronal network which comprises an essentially hierarchical network structure;
- the trainable artificial neuronal network, of which there is at least one, comprises an essentially hierarchical network structure which comprises at least two essentially independent, trainable artificial neuronal partial networks;
- the first artificial neuronal partial network comprises at least two independent artificial neuronal partial networks;
in each instance the results of the preliminary data processing operations are fed as input quantities to the first artificial neuronal partial networks;
- the results of the first artificial neuronal partial networks are fed as input quantities to the second artificial neuronal partial network;
and - the result of the artificial neuronal network, of which there is at least one, is used for quality control.
2. The method according to claim 1, characterised in that a sensor for acquiring the gap geometry is used, wherein the sensor is arranged in front of the weld location.
3. The method according to one of claims 1 to 2, characterised in that a sensor for acquiring the gap in the joint is used, wherein the sensor is arranged in front of the weld location.
4. The method according to one of claims 1 to 3, characterised in that a sensor for acquiring the pyro temperature is used, wherein the sensor is arranged just behind the weld location.
5. The method according to one of claims 1 to 4, characterised in that each artificial neuronal partial network comprises three layers, wherein the first layer comprises exactly one neuron, the second layer comprises a multitude of neurons, and the third layer comprises exactly one neuron.
6. The method according to one of claims 1 to 5, characterised in that the second artificial neuronal partial network comprises three layers, wherein the first layer comprises a multitude of neurons, the second layer comprises a multitude of neurons, and the third layer comprises exactly one neuron.
7. The method according to one of claims 1 to 6, characterised in that the learning process of the artificial neuronal network is carried out by means of a backpropagation learning algorithm in which the first artificial neuronal partial networks are adjusted at a learning rate .eta. between 0.01 and 0.1 and a momentum .alpha. between 0.1 and 0.6 and in which the second artificial neuronal partial network is adjusted at a learning rate .eta. and a momentum a which are essentially adapted to a gradient of an error function of the output of the artificial neuronal partial network.
8. The method according to one of claims 1 to 7, characterised in that a characteristics extraction of the sensor data is carried out during preliminary data processing.
9. The method according to claim 8, characterised in that during characteristics extraction of the sensor data characterising the gap width, the sensor data acquired by a sensor arranged at the weld location, is fed as an input quantity to an error suppression process, the results of the error suppression are fed as an input quantity to a window averaging process which essentially is freely definable, with the difference of the results of the window averaging process being established.
10. The method according to one of claims 8 or 9, characterised in that during characteristics extraction of the sensor data which characterises plasma intensity, the sensor data measured by a plasma intensity sensor can be fed as an input quantity to a window transformation.
11. The method according to one of claims 8 to 10 characterised in that during characteristics extraction of the sensor data characterising the weld sinkage, the sensor data which was measured by a geometry sensor is fed as an input quantity to a window averaging process.
12. The method according to one of claims 8 to 11 characterised in that during characteristics extraction of the sensor data which reflects misalignment of edges, the sensor data measured by a geometry sensor is fed as an input quantity to a mean value transformation process, and in that the result of this mean value transformation process is fed to a window averaging process.
13. The method according to one of claims 1 to 12 characterised in that the results of the first artificial neuronal partial networks are standardised in relation to the range of values, and are fed to the second artificial neuronal partial network.
14. A device, in particular to implement a method according to one of claims 1 to 13, characterised in that - at least two sensors for acquiring sensor data, are arranged around the weld location;
- ~wherein a sensor acquiring the weld plasma is arranged at the weld location; and - ~wherein a sensor acquiring the seam geometry is arranged after the weld location; and - ~in that the sensor data is used as an input quantity for one preliminary data processing setup each;
- ~in that the storage units store the results of preliminary data processing for the purpose of a co-locational relationship of the sensor data;
- in that the entries of the storage units are used as parallel input quantities for an essentially trainable artificial neuronal network structure;
- the trainable artificial neuronal network, of which there is at least one, comprises an essentially hierarchical network structure which comprises at least two essentially independent artificial neuronal partial networks;
- the first artificial neuronal partial network comprises at least two independent artificial neuronal partial networks;
- the first artificial neuronal partial networks each process the results of the preliminary data processing operations as input quantities;
- the second artificial neuronal partial networks process the results of the first artificial neuronal partial networks as input quantities;
and - the result of the neuronal network structure is used for qualitative assessment of the weld seam.
- ~wherein a sensor acquiring the weld plasma is arranged at the weld location; and - ~wherein a sensor acquiring the seam geometry is arranged after the weld location; and - ~in that the sensor data is used as an input quantity for one preliminary data processing setup each;
- ~in that the storage units store the results of preliminary data processing for the purpose of a co-locational relationship of the sensor data;
- in that the entries of the storage units are used as parallel input quantities for an essentially trainable artificial neuronal network structure;
- the trainable artificial neuronal network, of which there is at least one, comprises an essentially hierarchical network structure which comprises at least two essentially independent artificial neuronal partial networks;
- the first artificial neuronal partial network comprises at least two independent artificial neuronal partial networks;
- the first artificial neuronal partial networks each process the results of the preliminary data processing operations as input quantities;
- the second artificial neuronal partial networks process the results of the first artificial neuronal partial networks as input quantities;
and - the result of the neuronal network structure is used for qualitative assessment of the weld seam.
15. The device according to claim 14, characterised in that a geometry sensor is essentially arranged in front of the weld location.
16. The device according to one of claims 14 or 15, characterised in that a pyro sensor is arranged at the weld location.
17. The device according to one of claims 14 to 16, characterised in that a gap sensor is arranged in front of the weld location.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE19957163.5 | 1999-11-27 | ||
| DE19957163A DE19957163C1 (en) | 1999-11-27 | 1999-11-27 | Method and device for quality control of the seam on sheets or strips butt welded with a laser |
| PCT/EP2000/011109 WO2001039919A2 (en) | 1999-11-27 | 2000-11-10 | Method and device for quality control of the joint on sheets or strips butt-welded by means of a laser |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA2390873A1 true CA2390873A1 (en) | 2001-06-07 |
Family
ID=7930568
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002390873A Abandoned CA2390873A1 (en) | 1999-11-27 | 2000-11-10 | Method and device for quality control of the joint on sheets or strips butt-welded by means of a laser |
Country Status (7)
| Country | Link |
|---|---|
| EP (1) | EP1232036A2 (en) |
| JP (1) | JP2003516860A (en) |
| CN (1) | CN1384772A (en) |
| AU (1) | AU2505501A (en) |
| CA (1) | CA2390873A1 (en) |
| DE (1) | DE19957163C1 (en) |
| WO (1) | WO2001039919A2 (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9056368B2 (en) | 2008-11-21 | 2015-06-16 | Precitec Kg | Method and device for monitoring a laser processing operation to be performed on a workpiece and laser processing head having such a device |
| RU2635588C2 (en) * | 2013-03-29 | 2017-11-14 | Тойота Дзидося Кабусики Кайся | Welding section control device and method of its control |
| RU2635586C2 (en) * | 2013-04-15 | 2017-11-14 | Тойота Дзидося Кабусики Кайся | Welding section control device and method of its control with selecting area for evaporation glowing and thermal emission selection |
| RU2638140C2 (en) * | 2013-03-29 | 2017-12-11 | Тойота Дзидося Кабусики Кайся | Welding device with quality control of weld structure and welding method with quality control of weld structure |
| US11958135B2 (en) | 2018-08-24 | 2024-04-16 | Fanuc Corporation | Machining condition adjustment device and machine learning device |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6670574B1 (en) | 2002-07-31 | 2003-12-30 | Unitek Miyachi Corporation | Laser weld monitor |
| DE10304709B4 (en) * | 2003-02-06 | 2006-08-17 | Daimlerchrysler Ag | Method and apparatus for laser hybrid welding |
| JP4032045B2 (en) * | 2004-08-13 | 2008-01-16 | 新キャタピラー三菱株式会社 | DATA PROCESSING METHOD, DATA PROCESSING DEVICE, DIAGNOSIS METHOD, AND DIAGNOSIS DEVICE |
| DE102007059505A1 (en) | 2007-02-27 | 2008-10-02 | Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. | Quality control system for laser welding and cutting processes comprises laser tools, each connected to its own robot and to laser transmitting single beam to each tool, sensors determining process parameters and feeding data to processor |
| US20090200282A1 (en) * | 2008-02-08 | 2009-08-13 | Gm Global Technology Operations, Inc. | Weld signature monitoring method and apparatus |
| DE102008038332A1 (en) | 2008-08-19 | 2009-01-22 | Daimler Ag | Robotic welding process and assembly for e.g. automobile assembly uses module-logged electrical and optical welding parameters |
| DE102008058422A1 (en) * | 2008-11-21 | 2010-05-27 | Precitec Itm Gmbh | Monitoring laser machining process to be carried out on workpiece, comprises detecting actual measuring values by sensor, which monitors the laser machining process, and determining actual characteristic values from actual measuring values |
| FR2962568B1 (en) * | 2010-07-09 | 2012-08-17 | Renault Sa | PROCESS FOR CONTROLLING THE QUALITY OF A WELD |
| DE102012221218A1 (en) | 2011-11-22 | 2013-05-23 | Leibniz-Institut Für Festkörper- Und Werkstoffforschung Dresden E.V. | Device, useful for quality assurance of products manufactured by laser beam processing, includes laser beam processing apparatus, laser beam source, deflecting unit, and unit for determining and recording temperature at processing position |
| CN108039093A (en) * | 2013-03-11 | 2018-05-15 | 林肯环球股份有限公司 | External data is imported and analyzed using virtual reality welding system |
| JP6081833B2 (en) * | 2013-03-15 | 2017-02-15 | 株式会社総合車両製作所 | Laser welding method and laser welding apparatus |
| KR101780049B1 (en) | 2013-07-01 | 2017-09-19 | 한국전자통신연구원 | Apparatus and method for monitoring laser welding |
| US11934520B2 (en) * | 2018-03-28 | 2024-03-19 | Nvidia Corporation | Detecting data anomalies on a data interface using machine learning |
| CN109332937B (en) * | 2018-09-04 | 2021-06-08 | 广州大学 | Laser soldering quality detection method based on temperature curve |
| DE102018129425B4 (en) * | 2018-11-22 | 2020-07-30 | Precitec Gmbh & Co. Kg | System for recognizing a machining error for a laser machining system for machining a workpiece, laser machining system for machining a workpiece by means of a laser beam comprising the same, and method for detecting a machining error in a laser machining system for machining a workpiece |
| DE102018129441B4 (en) * | 2018-11-22 | 2023-11-16 | Precitec Gmbh & Co. Kg | System for monitoring a laser processing process, laser processing system and method for monitoring a laser processing process |
| CN109940305A (en) * | 2019-03-19 | 2019-06-28 | 天津大学 | Laser Deep Penetration Welding intelligent monitor system and method based on plasma electric signal |
| DE102019208037A1 (en) * | 2019-06-03 | 2020-12-03 | Robert Bosch Gmbh | Method and apparatus for improving a laser beam welding process |
| EP3885069A1 (en) | 2020-03-25 | 2021-09-29 | Bystronic Laser AG | Quality control of a laser machining process by means of machine learning |
| CN115413251A (en) * | 2020-03-25 | 2022-11-29 | 百超激光有限公司 | Quality control of laser machining process using machine learning |
| CN111814245B (en) * | 2020-09-02 | 2020-12-15 | 广东博智林机器人有限公司 | Ceiling joint layout acquisition method and device, electronic equipment and storage medium |
| CN115916451B (en) * | 2020-10-16 | 2024-02-20 | 百超激光有限公司 | Method, control unit and laser cutting system for calculating spatially and temporally resolved, combined setpoint data sets |
| DE102022200371A1 (en) | 2022-01-14 | 2023-07-20 | Robert Bosch Gesellschaft mit beschränkter Haftung | Process and diagnostic system for diagnosing faults in weld seams |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH04127984A (en) * | 1990-09-19 | 1992-04-28 | Hitachi Ltd | Method and device for laser welding |
| JPH04128962A (en) * | 1990-09-20 | 1992-04-30 | Fujitsu Ltd | Integrating system for neural network |
| JPH06344167A (en) * | 1993-06-07 | 1994-12-20 | Toshiba Corp | Laser beam machine |
| DE59402598D1 (en) * | 1993-11-30 | 1997-06-05 | Elpatronic Ag | Simultaneous temperature measurements on laser weld seams with at least two pyrometers and assignment to process parameters and seam quality |
| JP2885040B2 (en) * | 1993-12-27 | 1999-04-19 | 日産自動車株式会社 | Laser welding quality control method |
| JPH08206859A (en) * | 1995-02-07 | 1996-08-13 | Toshiba Corp | Device for deciding quality of laser welding |
| JPH10235490A (en) * | 1996-12-24 | 1998-09-08 | Kawasaki Steel Corp | Method for evaluating weld state of electric welding machine and device therefor |
| GB9719527D0 (en) * | 1997-09-12 | 1997-11-19 | Powerlasers Ltd | Fuzzy neural-network controller for laser welding systems |
-
1999
- 1999-11-27 DE DE19957163A patent/DE19957163C1/en not_active Expired - Fee Related
-
2000
- 2000-11-10 WO PCT/EP2000/011109 patent/WO2001039919A2/en not_active Application Discontinuation
- 2000-11-10 AU AU25055/01A patent/AU2505501A/en not_active Abandoned
- 2000-11-10 JP JP2001541641A patent/JP2003516860A/en active Pending
- 2000-11-10 CA CA002390873A patent/CA2390873A1/en not_active Abandoned
- 2000-11-10 CN CN00814905A patent/CN1384772A/en active Pending
- 2000-11-10 EP EP00988702A patent/EP1232036A2/en not_active Withdrawn
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9056368B2 (en) | 2008-11-21 | 2015-06-16 | Precitec Kg | Method and device for monitoring a laser processing operation to be performed on a workpiece and laser processing head having such a device |
| RU2635588C2 (en) * | 2013-03-29 | 2017-11-14 | Тойота Дзидося Кабусики Кайся | Welding section control device and method of its control |
| RU2638140C2 (en) * | 2013-03-29 | 2017-12-11 | Тойота Дзидося Кабусики Кайся | Welding device with quality control of weld structure and welding method with quality control of weld structure |
| RU2635586C2 (en) * | 2013-04-15 | 2017-11-14 | Тойота Дзидося Кабусики Кайся | Welding section control device and method of its control with selecting area for evaporation glowing and thermal emission selection |
| US11958135B2 (en) | 2018-08-24 | 2024-04-16 | Fanuc Corporation | Machining condition adjustment device and machine learning device |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2001039919A3 (en) | 2001-12-06 |
| JP2003516860A (en) | 2003-05-20 |
| AU2505501A (en) | 2001-06-12 |
| DE19957163C1 (en) | 2001-08-09 |
| WO2001039919A2 (en) | 2001-06-07 |
| EP1232036A2 (en) | 2002-08-21 |
| CN1384772A (en) | 2002-12-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2390873A1 (en) | Method and device for quality control of the joint on sheets or strips butt-welded by means of a laser | |
| Pal et al. | Artificial neural network modeling of weld joint strength prediction of a pulsed metal inert gas welding process using arc signals | |
| Kovacevic et al. | Neurofuzzy model-based weld fusion state estimation | |
| KR100503778B1 (en) | Weld quality measurement | |
| Renken et al. | Development of an adaptive, self-learning control concept for an additive manufacturing process | |
| US7640125B2 (en) | Method for controlling the quality of industrial processes and system therefrom | |
| Park et al. | Estimation of weld bead size in CO 2 laser welding by using multiple regression and neural network | |
| JP2019135638A (en) | Data driving method for automatically detecting abnormal workpiece during production process | |
| Lee et al. | Development of real-time diagnosis framework for angular misalignment of robot spot-welding system based on machine learning | |
| Aviles-Viñas et al. | Acquisition of welding skills in industrial robots | |
| EP1439441A2 (en) | Method for identification and management of errors and defects in assembly and welding lines for automotive vehicle bodies and body parts or the like and plant in accordance with said method | |
| CN108089457B (en) | Process quality control method based on-line finite element simulation | |
| Ma et al. | Mixed logical dynamical model for back bead width prediction of pulsed GTAW process with misalignment | |
| Piecuch et al. | Diagnostics of welding process based on thermovision images using convolutional neural network | |
| KR100498151B1 (en) | Method and device for precalculating previously unknown parameters of an industrial process | |
| EP4170448A1 (en) | Process signal reconstruction and anomaly detection in laser machining processes | |
| JP3312468B2 (en) | Body mounting posture measuring device and vehicle body gap step measuring device using the same | |
| JPH06110504A (en) | Method and device for estimating characteristic of process, method for monitoring process by using the estimating method and method for controlling process | |
| Rzeznik | In‐line laser measurement in the assembly process | |
| Boutin et al. | Development of a machine learning model for deviation from trajectory detection in multi-pass TIG welding in a narrow gap | |
| Wang et al. | Automatic process adjustment for reducing dimensional variation in discrete part machining processes | |
| Jurca | On-Line Monitor for the CO2-Laser Welding Process | |
| CN117490891A (en) | Quality monitoring method and device based on image processing | |
| Shouszhi et al. | The detection and compensation of tool wear in process | |
| DE102022112324A1 (en) | SYSTEMS, METHODS AND DEVICES FOR AN ARC WELDING PROCESS (AW) AND QUALITY CONTROL |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FZDE | Dead |