Abstract: The application status of fuzzy neural networks in welding process control, performance prediction and defect detection is reviewed. By combining the fuzzy inference system with the neural network technology, real-time control of the welding process can be realized, and an intelligent detection system for welding defects can be established to improve the intelligence level of the welding technology. Finally, the development trend of fuzzy neural network in welding is briefly introduced.
Key words: fuzzy neural network; process control; performance prediction; defect detection
With the rapid development of electronic technology, computer technology and artificial intelligence technology, the traditional mechanical manufacturing industry is undergoing a revolution. As an indispensable processing technology in the mechanical manufacturing industry, welding technology has also experienced tremendous impact from electronic technology, computer technology and artificial intelligence technology, which has made dramatic changes in welding process control, performance prediction and defect detection. .
Fuzzy neural network is a new technology that combines the advantages of fuzzy inference system and neural network. It has been widely used in the field of welding and is a hot topic in current research. Although fuzzy reasoning and neural network technology have certain application results in welding [1, 2], their inherent shortcomings limit the wide application in the field of welding.
1 Application of fuzzy neural network in welding process control
The welding process is a complex process with high nonlinearity and time-varying characteristics. It is difficult to obtain the best control effect by using the classical control theory to control the welding process to establish an accurate mathematical model. Fuzzy control absorbs the characteristics of human experience thinking, and neural network has the characteristics of self-organization and self-learning. The advantages and disadvantages of the two have obvious complementarity. The fuzzy neural network controller combining the two not only overcomes their shortcomings, but also receives extensive attention in the actual production control process.
1.1 Application of Fuzzy Neural Network Control Technology in Gas Shielded Arc Welding
1.1.1 Application of Fuzzy Neural Network Control Technology in GTAW
GTAW (Tungsten Gas Shielded Arc Welding) is a method for controlling line energy and performing high quality sheet welding. The GTAW process is a complex, multi-parameter, highly nonlinear system. It is difficult to achieve real-time and effective online control using traditional control methods in the actual welding process.
Literature [3] applied neural network-fuzzy control technology to study the modeling and control of GTAW process penetration. The image of the arc zone and the shape of the weld pool surface were obtained by the visual sensor CCD. The neural network adaptive resonance theory model algorithm was used to extract the melt width and weld gap information from the CCD image, and input the three-layer BP neural network together with the welding current. The amount of penetration, and then the difference between the actual penetration depth and the desired penetration depth ( ) and the rate of change of the difference ( ) are input to the fuzzy controller as control quantities to adjust the welding current, thereby achieving the purpose of adjusting the penetration depth. The penetration control system is shown in Figure 1. In the figure, U: current adjustment amount, G: weld gap, W: melt width, D: penetration depth.
Genetic algorithm is a kind of randomized search algorithm that draws on the natural selection and natural genetic mechanism of the biological world. It is suitable for dealing with complex and nonlinear problems that are difficult to solve by traditional search methods. It is one of the key technologies in intelligent computing.
On the basis of fuzzy neural network controller, the literature [4] designed a new type of fuzzy controller by using the advantages of fuzzy control, neural network and genetic algorithm. Based on the large amount of manual welding process data obtained in advance, the fuzzy neural network controller based on genetic algorithm was established by controlling the GTAW weld seam. Through the self-learning, training and timely feedback of the welding quality, the online monitoring welding quality was realized. Objective, the control effect is obtained, and the new fuzzy neural network controller has certain advantages over the traditional fuzzy controller.
1.1.2 Application of Fuzzy Neural Network Control Technology in Tungsten Arc Welding
TIG welding (TIG) can be used for all-position welding of almost any metal material. It is also an ideal method for double-sided forming of single-sided welding. It has been widely used in the manufacturing process of important product structures.
In [5], a fuzzy neural network control system is designed for the dynamic process control of pulsed TIG welding. The fuzzy neural network studied in this paper takes the current error as the input parameter and the current-adjusted fuzzy quantity as the output parameter, and uses the 3-layer BP neural network to adjust the fuzzy rule of the current. The pulsed TIG welding fusion width fuzzy neural network control system is shown in Fig. 2. Fuzzier's role is to limit and blur the error e into the input variable of the fuzzy neural network. ANN is a 3-layer BP network for memory fuzzy rules, and the output is refined by Defuzzier to obtain the actual value of the welding current adjustment. For the welding process, the MS is the inspection and image processing step.
Literature [6] used an improved fuzzy neural network controller to study the TIG back pool control. The front image of the molten pool is taken by the CCD camera, and the frontal geometrical parameters of the molten pool are extracted. The actual melting width of the back surface is calculated by using the relationship model between the front geometrical parameters of the molten pool and the backside melting width, and then the ideal melting width and the back surface are actually melted. The difference between the widths is used as the adjustment amount input fuzzy controller, and after the fuzzy rule inference, the adjustment amount of the welding current is output. The relationship between the geometrical parameters of the front surface of the molten pool and the backside melting width is expressed by a single-layer neural network. The input of this neural network is the front geometry of the molten pool and the adjusted welding current (ie, the original welding current plus the welding current adjustment). The output is the width of the back of the molten pool. In this paper, the fuzzy neural network controller with the above structure is compared with the change of the backside melt width when using the conventional PID controller, as shown in Figures 3 and 4.
It can be clearly seen from the figure that the fuzzy neural network controller has better control effect on the back width of the molten pool than the conventional PID controller. Simulation and experimental results show that the controller has good control performance and control effect.
1.1.3 Application of Fuzzy Neural Network Control Technology in MIG Welding
CO2 short-circuit transition welding is widely used in the welding of medium and thin steel structures. It is an efficient and energy-saving welding method, but it also has the disadvantages of large splash and poor forming. In [7], an adaptive fuzzy neural network control system for stabilizing CO2 short-circuit transition welding current is designed. The control system transforms the abstract empirical rules of fuzzy control into a set of input and output samples of the neural network, and uses the BP neural network training algorithm to train the pairs of learning samples. After that, for the input of the control process, the corresponding control amount can be output to achieve the purpose of controlling the current and voltage. Through software simulation and process test, it is proved that within the experimental current range, the maximum deviation of the average current does not exceed 7A, the average current relative error is less than 5%, and the deviation of the average current before control is not less than 12A, and the relative error is not less than 9%. The neuro-fuzzy controller eliminates welding current deviations due to random factors such as arc voltage regulation, network voltage fluctuations, shielding gas purity, flow rate, and torch height variation.
Literature [8] used the fuzzy neural network structure of neural network and fuzzy system to study the melting width and penetration control of stainless steel CO2 laser welding process. The input of the neural network after laser energy, scanning speed and plate thickness are processed by C-fuzzy clustering, and the welding efficiency (the ratio of input energy to molten metal volume), the melting width and the penetration depth as the output of the neural network, the neural network The output is input into the fuzzy inference system, and after fuzzy inference, a good or bad fuzzy evaluation is made. The experimental results show that the system can effectively control the CO2 laser welding process.
1.2 Application of Fuzzy Neural Network Control Technology in Resistance Spot Welding
The resistance spot welding process is a dynamic change process, especially the transient nature of the spot welding process and the invisibility of the weld nugget formation, which brings great difficulties to the control of the spot welding process quality. With the gradual maturity of intelligent control methods, the research work of introducing fuzzy neural network technology into the field of spot welding control is also increasing.
In [9], a spot welding quality control system combining neural network and fuzzy control technology was developed. The neural network module was used to map time to electrode displacement and electrode displacement rate, and the electrode displacement was used as the fuzzy deviation input of the fuzzy controller. The energy supplied to the nugget is taken as the output control amount. The computer simulation results show that the system has strong anti-interference ability, and can compensate online process disturbances, and can obtain satisfactory spot welding head.
In [10], the fuzzy neural network control technology is introduced into the resistance spot welding quality control system. The electrode displacement and its rate of change related to the displacement of the spot welding electrode without splashing and other parameters and different welding under the condition of splashing The dynamic resistance of the cycle is used as the input parameter of the fuzzy logic, and the fuzzy characteristic of the spot weld nugget is used as the output to construct a fuzzy system. The neural network learning algorithm is used as the pre-processing of fuzzy input, which improves the learning ability and intelligence of the system. The fuzzy reasoning uses Takagi-Sugeno as the kernel. The experimental results show that the system has a small prediction error, and the tensile shear strength of the resistance spot welded joint with and without splash can be inferred with higher efficiency.
1.3 Application of Fuzzy Neural Network Control Technology in Weld Seam Tracking
Weld seam tracking is one of the key technologies for welding automation in welding production. The early method adopted was to track the weld by arc sensor, eddy current sensor, ultrasonic sensor, and the like.
In [11], the CCD camera is used to capture the near-arc image. After the image processing, the deviation between the welding torch and the weld is calculated. The deviation and the deviation change rate are the input parameters of the network, and the correction amount is the output of the network. The correction amount calculated by the network is used as the correction amount of the robot swing arc welding, thereby achieving the effect of correcting the deviation. The experimental results show that it is feasible to control the deviation of the robot from the weld by the fuzzy neural network.
In [12], the fuzzy neural network is applied to the weld tracking system. The classical five-layer fuzzy neural network structure is adopted. The weld deviation and the deviation change rate are taken as input parameters, and the direction change of the welding torch is taken as the output parameter. Since the actual trajectory of the welding gun is biased to the target trajectory or the right deviation target trajectory, the output parameter of the network - the change direction of the welding gun direction is divided into five fuzzy membership degrees: {left bias, micro left bias, centering, micro right deviation, right deviation }. The experimental results show that the tracking system based on fuzzy neural network has a satisfactory tracking effect on the curved weld trajectory of straight weld and smooth transition.
In [13], a fuzzy control system for weld tracking neural network was designed. The CCD camera was used to obtain the photo of the weld pool and weld, and it was converted into a 256-level gray image and sent to the computer. A new method for extracting weld deviation and penetration information is presented in the literature. Weld deviation is the relative position difference between the torch and the weld. When extracting torch position information, according to the theory of heat transfer, the temperature of the arc center position (ie, the torch position) should be the highest, and the higher the temperature means the higher the gray level of the corresponding point on the thermal image. The author used this to successfully extract the position of the torch. The test results show that the control effect is good, and the new weld deviation information extraction method is superior to the traditional gradient method.
2 Application of fuzzy neural network in performance prediction of welded joints
Weld joints are prone to joint strength loss after welding and service, resulting in defects such as cracks and joint deformation. These problems often cause various damage accidents to the welded structure [14]. Therefore, it is necessary to predict the performance of welded joints.
In [15], the failure of beam-column joints commonly used in welded steel structures under earthquake action is analyzed, and the application of fuzzy neural network to the safety assessment of joints with defective beams and columns under seismic loading is made. the study. The results show that the fuzzy neural network method can evaluate the defects in the beam-column joints under the action of earthquakes; it can be used for the randomness, suddenness, location and importance of the steel structure and welds. The defects present in the process give an explicit treatment under random ground motion loads: rework, neglect, and disposal advice.
In [16], the neural network and adaptive neural network fuzzy inference toolbox provided by MATLAB were used to predict the weld crack opening displacement test data at low temperature. The input to the network is temperature and the output is the COD test data for the corresponding material. Experiments show that both neural network and adaptive neural network fuzzy inference system can obtain satisfactory results.
3 Application of Fuzzy Neural Network in Defect Detection of Welded Joints
Welding defect inspection is an important part of ensuring the quality of welding products. In the past, the detection of welding defects was carried out by an X-ray flaw detector, and then the quality of the welding was judged by an experienced professional film reviewer by visual inspection. The efficiency was low and the inspection level was unstable due to the subjective factors of the examiner. With the development of computer technology, image processing and artificial intelligence technology, it has become possible to use computers to conduct intelligent evaluation of defects.
Literature [17] established a computer-aided identification system that uses ray negatives to identify different types of welding defects. Firstly, the defect image is separated from the background by image processing technology, and then the fuzzy neural network of K-Nearest Neighbor algorithm is used to identify the defect type and compare it with the statistical self-help method (bootstrap).
Literature [18] conducted a study on the identification of real-time radiographic film weld defects in fuzzy neural networks. Firstly, the original image is denoised and processed, and the characteristic parameters such as the perimeter and area ratio of the defect, the inclination angle of the weld, the position of the centroid coordinate relative to the center of the weld, and the relative gray parameter are used as input parameters of the defect recognition network, and The contour tracking method is used to extract the characteristics of various welding defects, and the fuzzy neural network is used for identification analysis. The recognition results are given by the defect code. After training the model with 52 typical defect samples, 8 defect samples were subjected to recognition tests. The test results show that the method can effectively identify the weld defects and the effect is better than the traditional classification identification method.
Literature [19] proposed a fuzzy evaluation method based on BP neural network for the influence of welding defects on structural fatigue performance, which makes up for the difficulty in extracting the defect assessment rules when using fuzzy reasoning method to judge the influence of welding defects on structural fatigue strength. Rely on strong issues. At the time of judging, the defect type, the defect size, the relative position of the defect and the interference between the defects are taken as input parameters, and the degree of damage to the fatigue performance of the defect is taken as the output parameter. The neural network is trained using 40 sets of data, and the network prediction result is consistent with the experimental result. It is indicated that the evaluation method adopted in the literature is feasible.
4 Development trends
At present, the research and application of fuzzy neural network technology in welding process control, performance prediction and defect detection has made great progress, and has achieved gratifying results. However, due to the slow progress of fuzzy neural network theory and model algorithm research, and the overall backward level of domestic welding, the application level of fuzzy neural network in welding has yet to be further deepened and improved. The main trends for future development are as follows: (1) The fuzzy neural network currently used in the welding field, its neural network model is mainly BP model, it is necessary to study the fuzzy reasoning technology combined with other neural network models and network training algorithms in welding Application, establish a new and more reasonable model; (2) further expand the application of fuzzy neural network in welding, such as the application of fuzzy neural network to the prediction of welded joint defects; (3) the fuzzy neural network technology Combined with computer software technology, the software of commercial fuzzy neural network in welding process control and performance prediction is introduced to accelerate the development of domestic welding intelligent technology.
literature
[1] Wang Yu, Gao Dalu, Zhang Guang. The application and development trend of new artificial intelligence technology in welding [J]. Mechanical Science and Technology, 2002, 21(3): 494-496
[2] Song Dongfeng, Hu Shengyu. The development of fuzzy control technology and its application in the field of welding [J]. Electric Welder, 2005, 35(8): 23-25
[3] Gao Xiangdong, Huang Shisheng, Wu Naiyou. Research on GTAW neural network-fuzzy control technology [J]. Journal of Welding, 2000, 21(1) 5-8
[4] Lei Yucheng, Zhang Cheng, Cheng Xiaonong. Application of fuzzy neural network controller based on genetic algorithm in GTAW [J]. Journal of Welding, 2003, 24(4): 47-49
[5] Chen Shanben, Wu Lin, Wang Qilong and so on. Pulsed TIG welding melting width dynamic process fuzzy reasoning-neural network control method [J]. Journal of Welding, 1997, 18(3): 159-164
[6] Gao Jinqiang, Wu Chuansong, Liu Xinfeng. Neural network fuzzy control of TIG welding backside melting width [J]. Journal of Welding, 2001, 22(5): 5
[7] Wang Yasheng, Dai Fenglei, Cai Benhua. Carbon dioxide short circuit transition welding current neural fuzzy control system [J]. Journal of Xi'an Jiaotong University, 2003, 37(3): 283-285
[8] G Casalino, F Memola Capece Minutolo. A model for evaluation of laser welding efficiency and quality using an artificial neural network and fuzzy logic [J]. J.engineering Manufacture,2004,218:641
[9]Jou, Min. An Intelligent Control System for Resistance Spot Welding Using Fuzzy Logic and Neural Network [J]. Rensselaer Polytechnic Institute. 1994
[10] Lee SR, Choo YJ, Lee TY, et al. A quality assurance technique for resistance spot welding using a neuro-fuzzy algorithm [J], Journal of Manufacturing Systems, 2001, (5): 320~328
[11] Wang Gang, Xu Ying, Zhang Chuanying. Fuzzy neural network control system for robot swing arc welding correction [J]. Basic Automation, 2000, 7(4): 17-19
[12] Sun Hua, Xu Songyuan, Jiang Yanzhu. Weld seam tracking algorithm based on fuzzy neural network [J]. Journal of Harbin University of Science and Technology, 2001, 6(4): 67-70
[13] Zhang Hua, Hu Jing, Wrink Chunhua, etc. Weld deviation and penetration identification and its integrated intelligent control system [J]. Journal of Welding, 2003, 24(4): 51-54
[14] printed with victory. Metal welding defects and their prevention [M]. Harbin: Heilongjiang Science and Technology Press. 1995: 574-633
[15] Li Jie, Zhang Yufeng and so on. Reliability fuzzy evaluation of beam-column joints under seismic loading [C]. Proceedings of the Tenth National Welding Conference. 2001 (2): 291-293
[16] Liu Changhong, Chen Qiu. Discussion on prediction method of low temperature crack opening displacement [J]. Welding Technology. 2005, 34(4): 7-8
[17] Gang Wang, T. Warren Liao. Automatic identification of different types of welding defects in radiographic images [J]. NDT&E International, 2002, 35: 519-520
[18] Zhang Xiaoguang, Lin Jiajun. Research on Weld Defect Recognition Method Based on Fuzzy Neural Network [J]. Journal of China University of Mining and Technology, 2003, 32(1): 91-94
[19] Yu Shurong, He Shiquan, Li Erguo, etc. Intelligent comprehensive evaluation of the influence of welding defects on structural fatigue performance [J]. Petrochemical equipment, 2001, 30 (3): 1-4
Key words: fuzzy neural network; process control; performance prediction; defect detection
With the rapid development of electronic technology, computer technology and artificial intelligence technology, the traditional mechanical manufacturing industry is undergoing a revolution. As an indispensable processing technology in the mechanical manufacturing industry, welding technology has also experienced tremendous impact from electronic technology, computer technology and artificial intelligence technology, which has made dramatic changes in welding process control, performance prediction and defect detection. .
Fuzzy neural network is a new technology that combines the advantages of fuzzy inference system and neural network. It has been widely used in the field of welding and is a hot topic in current research. Although fuzzy reasoning and neural network technology have certain application results in welding [1, 2], their inherent shortcomings limit the wide application in the field of welding.
1 Application of fuzzy neural network in welding process control
The welding process is a complex process with high nonlinearity and time-varying characteristics. It is difficult to obtain the best control effect by using the classical control theory to control the welding process to establish an accurate mathematical model. Fuzzy control absorbs the characteristics of human experience thinking, and neural network has the characteristics of self-organization and self-learning. The advantages and disadvantages of the two have obvious complementarity. The fuzzy neural network controller combining the two not only overcomes their shortcomings, but also receives extensive attention in the actual production control process.
1.1 Application of Fuzzy Neural Network Control Technology in Gas Shielded Arc Welding
1.1.1 Application of Fuzzy Neural Network Control Technology in GTAW
GTAW (Tungsten Gas Shielded Arc Welding) is a method for controlling line energy and performing high quality sheet welding. The GTAW process is a complex, multi-parameter, highly nonlinear system. It is difficult to achieve real-time and effective online control using traditional control methods in the actual welding process.
Literature [3] applied neural network-fuzzy control technology to study the modeling and control of GTAW process penetration. The image of the arc zone and the shape of the weld pool surface were obtained by the visual sensor CCD. The neural network adaptive resonance theory model algorithm was used to extract the melt width and weld gap information from the CCD image, and input the three-layer BP neural network together with the welding current. The amount of penetration, and then the difference between the actual penetration depth and the desired penetration depth ( ) and the rate of change of the difference ( ) are input to the fuzzy controller as control quantities to adjust the welding current, thereby achieving the purpose of adjusting the penetration depth. The penetration control system is shown in Figure 1. In the figure, U: current adjustment amount, G: weld gap, W: melt width, D: penetration depth.
Genetic algorithm is a kind of randomized search algorithm that draws on the natural selection and natural genetic mechanism of the biological world. It is suitable for dealing with complex and nonlinear problems that are difficult to solve by traditional search methods. It is one of the key technologies in intelligent computing.
On the basis of fuzzy neural network controller, the literature [4] designed a new type of fuzzy controller by using the advantages of fuzzy control, neural network and genetic algorithm. Based on the large amount of manual welding process data obtained in advance, the fuzzy neural network controller based on genetic algorithm was established by controlling the GTAW weld seam. Through the self-learning, training and timely feedback of the welding quality, the online monitoring welding quality was realized. Objective, the control effect is obtained, and the new fuzzy neural network controller has certain advantages over the traditional fuzzy controller.
1.1.2 Application of Fuzzy Neural Network Control Technology in Tungsten Arc Welding
TIG welding (TIG) can be used for all-position welding of almost any metal material. It is also an ideal method for double-sided forming of single-sided welding. It has been widely used in the manufacturing process of important product structures.
In [5], a fuzzy neural network control system is designed for the dynamic process control of pulsed TIG welding. The fuzzy neural network studied in this paper takes the current error as the input parameter and the current-adjusted fuzzy quantity as the output parameter, and uses the 3-layer BP neural network to adjust the fuzzy rule of the current. The pulsed TIG welding fusion width fuzzy neural network control system is shown in Fig. 2. Fuzzier's role is to limit and blur the error e into the input variable of the fuzzy neural network. ANN is a 3-layer BP network for memory fuzzy rules, and the output is refined by Defuzzier to obtain the actual value of the welding current adjustment. For the welding process, the MS is the inspection and image processing step.
Literature [6] used an improved fuzzy neural network controller to study the TIG back pool control. The front image of the molten pool is taken by the CCD camera, and the frontal geometrical parameters of the molten pool are extracted. The actual melting width of the back surface is calculated by using the relationship model between the front geometrical parameters of the molten pool and the backside melting width, and then the ideal melting width and the back surface are actually melted. The difference between the widths is used as the adjustment amount input fuzzy controller, and after the fuzzy rule inference, the adjustment amount of the welding current is output. The relationship between the geometrical parameters of the front surface of the molten pool and the backside melting width is expressed by a single-layer neural network. The input of this neural network is the front geometry of the molten pool and the adjusted welding current (ie, the original welding current plus the welding current adjustment). The output is the width of the back of the molten pool. In this paper, the fuzzy neural network controller with the above structure is compared with the change of the backside melt width when using the conventional PID controller, as shown in Figures 3 and 4.
It can be clearly seen from the figure that the fuzzy neural network controller has better control effect on the back width of the molten pool than the conventional PID controller. Simulation and experimental results show that the controller has good control performance and control effect.
1.1.3 Application of Fuzzy Neural Network Control Technology in MIG Welding
CO2 short-circuit transition welding is widely used in the welding of medium and thin steel structures. It is an efficient and energy-saving welding method, but it also has the disadvantages of large splash and poor forming. In [7], an adaptive fuzzy neural network control system for stabilizing CO2 short-circuit transition welding current is designed. The control system transforms the abstract empirical rules of fuzzy control into a set of input and output samples of the neural network, and uses the BP neural network training algorithm to train the pairs of learning samples. After that, for the input of the control process, the corresponding control amount can be output to achieve the purpose of controlling the current and voltage. Through software simulation and process test, it is proved that within the experimental current range, the maximum deviation of the average current does not exceed 7A, the average current relative error is less than 5%, and the deviation of the average current before control is not less than 12A, and the relative error is not less than 9%. The neuro-fuzzy controller eliminates welding current deviations due to random factors such as arc voltage regulation, network voltage fluctuations, shielding gas purity, flow rate, and torch height variation.
Literature [8] used the fuzzy neural network structure of neural network and fuzzy system to study the melting width and penetration control of stainless steel CO2 laser welding process. The input of the neural network after laser energy, scanning speed and plate thickness are processed by C-fuzzy clustering, and the welding efficiency (the ratio of input energy to molten metal volume), the melting width and the penetration depth as the output of the neural network, the neural network The output is input into the fuzzy inference system, and after fuzzy inference, a good or bad fuzzy evaluation is made. The experimental results show that the system can effectively control the CO2 laser welding process.
1.2 Application of Fuzzy Neural Network Control Technology in Resistance Spot Welding
The resistance spot welding process is a dynamic change process, especially the transient nature of the spot welding process and the invisibility of the weld nugget formation, which brings great difficulties to the control of the spot welding process quality. With the gradual maturity of intelligent control methods, the research work of introducing fuzzy neural network technology into the field of spot welding control is also increasing.
In [9], a spot welding quality control system combining neural network and fuzzy control technology was developed. The neural network module was used to map time to electrode displacement and electrode displacement rate, and the electrode displacement was used as the fuzzy deviation input of the fuzzy controller. The energy supplied to the nugget is taken as the output control amount. The computer simulation results show that the system has strong anti-interference ability, and can compensate online process disturbances, and can obtain satisfactory spot welding head.
In [10], the fuzzy neural network control technology is introduced into the resistance spot welding quality control system. The electrode displacement and its rate of change related to the displacement of the spot welding electrode without splashing and other parameters and different welding under the condition of splashing The dynamic resistance of the cycle is used as the input parameter of the fuzzy logic, and the fuzzy characteristic of the spot weld nugget is used as the output to construct a fuzzy system. The neural network learning algorithm is used as the pre-processing of fuzzy input, which improves the learning ability and intelligence of the system. The fuzzy reasoning uses Takagi-Sugeno as the kernel. The experimental results show that the system has a small prediction error, and the tensile shear strength of the resistance spot welded joint with and without splash can be inferred with higher efficiency.
1.3 Application of Fuzzy Neural Network Control Technology in Weld Seam Tracking
Weld seam tracking is one of the key technologies for welding automation in welding production. The early method adopted was to track the weld by arc sensor, eddy current sensor, ultrasonic sensor, and the like.
In [11], the CCD camera is used to capture the near-arc image. After the image processing, the deviation between the welding torch and the weld is calculated. The deviation and the deviation change rate are the input parameters of the network, and the correction amount is the output of the network. The correction amount calculated by the network is used as the correction amount of the robot swing arc welding, thereby achieving the effect of correcting the deviation. The experimental results show that it is feasible to control the deviation of the robot from the weld by the fuzzy neural network.
In [12], the fuzzy neural network is applied to the weld tracking system. The classical five-layer fuzzy neural network structure is adopted. The weld deviation and the deviation change rate are taken as input parameters, and the direction change of the welding torch is taken as the output parameter. Since the actual trajectory of the welding gun is biased to the target trajectory or the right deviation target trajectory, the output parameter of the network - the change direction of the welding gun direction is divided into five fuzzy membership degrees: {left bias, micro left bias, centering, micro right deviation, right deviation }. The experimental results show that the tracking system based on fuzzy neural network has a satisfactory tracking effect on the curved weld trajectory of straight weld and smooth transition.
In [13], a fuzzy control system for weld tracking neural network was designed. The CCD camera was used to obtain the photo of the weld pool and weld, and it was converted into a 256-level gray image and sent to the computer. A new method for extracting weld deviation and penetration information is presented in the literature. Weld deviation is the relative position difference between the torch and the weld. When extracting torch position information, according to the theory of heat transfer, the temperature of the arc center position (ie, the torch position) should be the highest, and the higher the temperature means the higher the gray level of the corresponding point on the thermal image. The author used this to successfully extract the position of the torch. The test results show that the control effect is good, and the new weld deviation information extraction method is superior to the traditional gradient method.
2 Application of fuzzy neural network in performance prediction of welded joints
Weld joints are prone to joint strength loss after welding and service, resulting in defects such as cracks and joint deformation. These problems often cause various damage accidents to the welded structure [14]. Therefore, it is necessary to predict the performance of welded joints.
In [15], the failure of beam-column joints commonly used in welded steel structures under earthquake action is analyzed, and the application of fuzzy neural network to the safety assessment of joints with defective beams and columns under seismic loading is made. the study. The results show that the fuzzy neural network method can evaluate the defects in the beam-column joints under the action of earthquakes; it can be used for the randomness, suddenness, location and importance of the steel structure and welds. The defects present in the process give an explicit treatment under random ground motion loads: rework, neglect, and disposal advice.
In [16], the neural network and adaptive neural network fuzzy inference toolbox provided by MATLAB were used to predict the weld crack opening displacement test data at low temperature. The input to the network is temperature and the output is the COD test data for the corresponding material. Experiments show that both neural network and adaptive neural network fuzzy inference system can obtain satisfactory results.
3 Application of Fuzzy Neural Network in Defect Detection of Welded Joints
Welding defect inspection is an important part of ensuring the quality of welding products. In the past, the detection of welding defects was carried out by an X-ray flaw detector, and then the quality of the welding was judged by an experienced professional film reviewer by visual inspection. The efficiency was low and the inspection level was unstable due to the subjective factors of the examiner. With the development of computer technology, image processing and artificial intelligence technology, it has become possible to use computers to conduct intelligent evaluation of defects.
Literature [17] established a computer-aided identification system that uses ray negatives to identify different types of welding defects. Firstly, the defect image is separated from the background by image processing technology, and then the fuzzy neural network of K-Nearest Neighbor algorithm is used to identify the defect type and compare it with the statistical self-help method (bootstrap).
Literature [18] conducted a study on the identification of real-time radiographic film weld defects in fuzzy neural networks. Firstly, the original image is denoised and processed, and the characteristic parameters such as the perimeter and area ratio of the defect, the inclination angle of the weld, the position of the centroid coordinate relative to the center of the weld, and the relative gray parameter are used as input parameters of the defect recognition network, and The contour tracking method is used to extract the characteristics of various welding defects, and the fuzzy neural network is used for identification analysis. The recognition results are given by the defect code. After training the model with 52 typical defect samples, 8 defect samples were subjected to recognition tests. The test results show that the method can effectively identify the weld defects and the effect is better than the traditional classification identification method.
Literature [19] proposed a fuzzy evaluation method based on BP neural network for the influence of welding defects on structural fatigue performance, which makes up for the difficulty in extracting the defect assessment rules when using fuzzy reasoning method to judge the influence of welding defects on structural fatigue strength. Rely on strong issues. At the time of judging, the defect type, the defect size, the relative position of the defect and the interference between the defects are taken as input parameters, and the degree of damage to the fatigue performance of the defect is taken as the output parameter. The neural network is trained using 40 sets of data, and the network prediction result is consistent with the experimental result. It is indicated that the evaluation method adopted in the literature is feasible.
4 Development trends
At present, the research and application of fuzzy neural network technology in welding process control, performance prediction and defect detection has made great progress, and has achieved gratifying results. However, due to the slow progress of fuzzy neural network theory and model algorithm research, and the overall backward level of domestic welding, the application level of fuzzy neural network in welding has yet to be further deepened and improved. The main trends for future development are as follows: (1) The fuzzy neural network currently used in the welding field, its neural network model is mainly BP model, it is necessary to study the fuzzy reasoning technology combined with other neural network models and network training algorithms in welding Application, establish a new and more reasonable model; (2) further expand the application of fuzzy neural network in welding, such as the application of fuzzy neural network to the prediction of welded joint defects; (3) the fuzzy neural network technology Combined with computer software technology, the software of commercial fuzzy neural network in welding process control and performance prediction is introduced to accelerate the development of domestic welding intelligent technology.
literature
[1] Wang Yu, Gao Dalu, Zhang Guang. The application and development trend of new artificial intelligence technology in welding [J]. Mechanical Science and Technology, 2002, 21(3): 494-496
[2] Song Dongfeng, Hu Shengyu. The development of fuzzy control technology and its application in the field of welding [J]. Electric Welder, 2005, 35(8): 23-25
[3] Gao Xiangdong, Huang Shisheng, Wu Naiyou. Research on GTAW neural network-fuzzy control technology [J]. Journal of Welding, 2000, 21(1) 5-8
[4] Lei Yucheng, Zhang Cheng, Cheng Xiaonong. Application of fuzzy neural network controller based on genetic algorithm in GTAW [J]. Journal of Welding, 2003, 24(4): 47-49
[5] Chen Shanben, Wu Lin, Wang Qilong and so on. Pulsed TIG welding melting width dynamic process fuzzy reasoning-neural network control method [J]. Journal of Welding, 1997, 18(3): 159-164
[6] Gao Jinqiang, Wu Chuansong, Liu Xinfeng. Neural network fuzzy control of TIG welding backside melting width [J]. Journal of Welding, 2001, 22(5): 5
[7] Wang Yasheng, Dai Fenglei, Cai Benhua. Carbon dioxide short circuit transition welding current neural fuzzy control system [J]. Journal of Xi'an Jiaotong University, 2003, 37(3): 283-285
[8] G Casalino, F Memola Capece Minutolo. A model for evaluation of laser welding efficiency and quality using an artificial neural network and fuzzy logic [J]. J.engineering Manufacture,2004,218:641
[9]Jou, Min. An Intelligent Control System for Resistance Spot Welding Using Fuzzy Logic and Neural Network [J]. Rensselaer Polytechnic Institute. 1994
[10] Lee SR, Choo YJ, Lee TY, et al. A quality assurance technique for resistance spot welding using a neuro-fuzzy algorithm [J], Journal of Manufacturing Systems, 2001, (5): 320~328
[11] Wang Gang, Xu Ying, Zhang Chuanying. Fuzzy neural network control system for robot swing arc welding correction [J]. Basic Automation, 2000, 7(4): 17-19
[12] Sun Hua, Xu Songyuan, Jiang Yanzhu. Weld seam tracking algorithm based on fuzzy neural network [J]. Journal of Harbin University of Science and Technology, 2001, 6(4): 67-70
[13] Zhang Hua, Hu Jing, Wrink Chunhua, etc. Weld deviation and penetration identification and its integrated intelligent control system [J]. Journal of Welding, 2003, 24(4): 51-54
[14] printed with victory. Metal welding defects and their prevention [M]. Harbin: Heilongjiang Science and Technology Press. 1995: 574-633
[15] Li Jie, Zhang Yufeng and so on. Reliability fuzzy evaluation of beam-column joints under seismic loading [C]. Proceedings of the Tenth National Welding Conference. 2001 (2): 291-293
[16] Liu Changhong, Chen Qiu. Discussion on prediction method of low temperature crack opening displacement [J]. Welding Technology. 2005, 34(4): 7-8
[17] Gang Wang, T. Warren Liao. Automatic identification of different types of welding defects in radiographic images [J]. NDT&E International, 2002, 35: 519-520
[18] Zhang Xiaoguang, Lin Jiajun. Research on Weld Defect Recognition Method Based on Fuzzy Neural Network [J]. Journal of China University of Mining and Technology, 2003, 32(1): 91-94
[19] Yu Shurong, He Shiquan, Li Erguo, etc. Intelligent comprehensive evaluation of the influence of welding defects on structural fatigue performance [J]. Petrochemical equipment, 2001, 30 (3): 1-4
Galvanized Roller Shutter Door
Using advanced hot-dip galvanized material , precision roll forming production process, polyurethane foam that free of hydrofluorocarbon inside door, light weight, high strength, appearance nice. The doors have advantages of tamper anti-theft , noise isolation thermal insulation and corrosion resistance;the doors are widely used in shops, garages, houses, supermarkets, warehouses and industrial plants,etc.
Depending on the characteristics of the buildings,we can choose interior, exterior and intermediate three different installation ways, doors stay in the upper part of door after rolling up , this could save garage space, and also could keep the garage clean and beautiful by prevent the mud, rain and snow ,etc going into the garage.
Galvanized Roller Shutter Door,Standard Galvanized Roller Shutter Door,Automatic Galvanized Roller Shutter Door,Steel Roller Shutter Safe Door
Shenzhen Hongfa Automatic Door Co., Ltd. , https://www.hfgaragedoor.com