I. Introduction
Welding robots are the most widely used type of industrial robot, accounting for approximately 40% to 60% of the total robot applications in various countries. The adoption of robotic welding represents a revolutionary advancement in welding automation, breaking through traditional rigid automation methods and pioneering a new approach of flexible automation. Rigid automated welding equipment is generally specialized and typically used for the automated production of medium to large batches of welded products. Therefore, in the welding production of medium to small batches of products, shielded metal arc welding remains the primary welding method. Welding robots make the automated welding production of small batches of products possible.
The main advantages of welding robots are as follows:
1) It is easy to achieve stability and improvement in the quality of welded products, ensuring their uniformity;
2) Increase productivity, enabling continuous 24-hour production;
3) Improved working conditions for workers, enabling them to work long hours in hazardous environments:
4) Reduce the technical skill requirements for workers;
5) Shorten the preparation cycle for product upgrades and replacements, and reduce the corresponding equipment investment;
6) It can automate the welding of small batches of products;
7) Provides a technological foundation for flexible welding production lines.
II. Technical Components of Intelligent Robotic Welding Systems
Intelligent robotic welding systems are advanced manufacturing systems built upon intelligent feedback control theory, involving the integration of numerous disciplines and technologies. Besides the different welding process requirements necessitating varying welding robot technologies and related equipment, current intelligent robotic welding systems can be broadly categorized into the components shown in Figure 1:
1) Design technology for robot welding task planning software systems;
2) Intelligent sensing technology for welding environment, weld location and orientation, and welding dynamic process;
3) Robot motion trajectory control technology;
4) Design of a real-time intelligent controller for the dynamic welding process;
5) Control and optimization management technology for intelligent complex robotic welding systems.
III. Robot Welding Task Planning Technology
The basic task of the robot welding task planning system is to automatically generate the robot motion sequence from the initial state to the target state, the achievable welding torch motion trajectory and the optimal welding torch posture, as well as the matching welding parameters and control program within a certain welding work area, and to realize the automatic simulation and optimization of the welding planning process.
Robotic welding task planning can be categorized as a problem-solving technique within the field of artificial intelligence, comprising two parts: welding path planning and welding parameter planning. Due to the diversity and complexity of welding processes and tasks, it is essential to conduct computer software planning (i.e., CAD simulation design research) of the robotic welding path and welding parameter scheme before actual welding. This significantly reduces the time spent on production line teaching, improving the utilization rate of welding robots. Furthermore, it enables pre-welding simulation of the robot's motion process, ensuring the effectiveness and safety of the production process.
Robotic welding parameter planning mainly refers to the design and determination of various quality control parameters during the welding process. The foundation of welding parameter planning is the establishment of a parameter planning model. Due to the complexity and uncertainty of the welding process, the most commonly used and researched model structures are based on neural network theory, fuzzy inference theory, and expert system theory. Based on the model's structure and input-output relationship, the input and target parameters required by the parameter planning model can be generated from pre-acquired weld feature point data. After passing through the planner, the corresponding welding process parameters for welding can be obtained.
Robotic welding path planning differs from that of general mobile robots. Its key feature lies in the comprehensive design and optimization of the continuous curved trajectory of the weld seam, collision-free welding torch movement, path, and torch posture. Since welding parameter planning typically requires adjustments based on different process requirements, weld seam locations, and workpiece materials and shapes, and welding path planning and parameter planning are interconnected, joint planning research between them is of practical significance. For welding quality, the torch posture and welding parameters form a tightly coupled unified whole. On one hand, the torch posture in robot path planning determines the travel angle and working angle during welding, and the speed of the robot's end effector also determines the welding speed; travel angle, working angle, and welding speed are all important components of welding parameters. On the other hand, from the perspective of welding process and welding quality control, adjustments to parameters such as welding speed and torch travel angle must be implemented in the robot's motion path planning. Furthermore, from the perspective of weld seam formation planning models, the four quantities—welding current, arc voltage, torch speed, and welding travel angle—must be organically coordinated to achieve effective control over weld seam formation. Therefore, welding path and welding parameters are an organic and unified whole, and joint planning of welding path and welding parameters is necessary.
IV. Robot Welding Sensing Technology
One of the hallmarks of human intelligence is the ability to perceive the external world and take adaptive actions based on perceived information. To imbue robotic welding systems with a certain level of intelligence, it is essential to research intelligent sensing technologies for the robot's perception of the welding environment, weld position and direction, and the dynamic welding process. The robot's ability to perceive the welding environment can be achieved using computational and visual technologies. A visual model of the overall or partial environment of the workpiece can be used as the basis for planning welding tasks, collision-free paths, and welding parameters. This requires establishing a 3D vision hardware system and implementing image understanding, object segmentation, and recognition algorithm software.
Visual weld seam tracking sensors are one of the core and fundamental components of welding robot sensing systems. To acquire the three-dimensional contour of the weld joint and overcome interference from arc light during welding, robot weld seam tracking and recognition technology generally employs active vision methods such as lasers and structured light to correctly guide the robot's welding torch tip along the actual weld seam to complete the desired trajectory. Since the energy of the active light sources used is generally lower than that of the arc light, these sensors are typically placed at the front end of the welding torch to avoid interference from direct arc light. Active light sources are generally single- or multi-surface laser beams with scanning laser domains, and their processing is stable, simple, and practical.
Structured light vision is another form of active vision-based weld seam tracking. The corresponding sensor mainly consists of two parts: a projector that uses its radiant energy to form a projection surface, and a photoelectric position detection device, often a CCD camera with an area array. These components are assembled in a specific positional relationship and coupled with a certain algorithm to form a structured light vision sensor, capable of sensing the three-dimensional information of all visible points on the projection surface. The trajectory of a spatial weld seam can be viewed as being composed of a series of discrete points, the density of which depends on the control requirements. The origin of the weld seam coordinate system is established at these points. The sensor measures the pose of one weld seam point at a time and can obtain heuristic information about the pose of unknown weld seam points. The guided robot's welding torch completes the tracking of the entire smooth, continuous weld seam.
Real-time detection technology for welding dynamic processes mainly refers to the online detection of parameters such as molten pool size, penetration, forming, and arc-induced behavior during the welding process, thereby achieving real-time control of welding quality. Due to the complex physicochemical reactions, strong nonlinearity, and numerous uncertainties inherent in the welding process, reliable and practical detection of the welding process has become a significant challenge. For a long time, many scholars have explored various approaches and technologies for detection, achieving success under certain conditions. These diverse detection methods, information processing techniques, and different sensing principles and technical implementation methods essentially require an improvement in comprehensive technology. From the perspective of detecting dynamic changes in the molten pool and penetration characteristics, computer vision technology, temperature field measurement, molten pool excitation oscillation, and arc sensing are currently considered to be effective methods for real-time control.
V. Intelligent Control Technology for Welding Dynamic Process
The welding dynamic process is a complex process influenced by multiple factors. In particular, the real-time control of the weld pool dimensions (i.e., weld width, weld depth, penetration, and weld formation) during arc welding is a significant challenge. Due to the strong nonlinearity of the controlled object, the coupling of multiple variables, the complexity of material physicochemical changes, and the presence of numerous random disturbances and uncertainties, effective real-time control of welding quality has been a long-standing and prominent issue in the welding field. It is also a crucial problem that cannot be overcome in realizing intelligent welding robot systems.
Since the controller design methods provided by classical and modern control theories are based on the precise mathematical model of the controlled object, and the welding dynamic process cannot provide such a controllable mathematical model, it is difficult to apply these theoretical methods to design an effective controller for the welding process.
In recent years, the emergence of intelligent control theories and methods such as fuzzy logic, artificial neural networks, and expert systems, which simulate human intelligent behavior, has made it possible to design intelligent controllers that simulate welder operations, aiming to solve the problem of real-time welding quality control. Currently, some scholars have integrated artificial intelligence technologies such as fuzzy logic, artificial neural networks, and expert reasoning into the dynamic process control of welding in robotic systems.
Designing intelligent controllers for actual welding dynamic process control requires considerable skill, especially in the research and system implementation of real-time adaptive and self-learning algorithms. Furthermore, different welding processes and detection methods will necessitate different intelligent controller design approaches. Combining intelligent controllers for welding dynamic processes with welding robot system design will substantially improve the intelligent technology of robotic welding.
VI. Intelligent Integrated System for Robotic Welding
For complex systems centered on welding robots, comprising welding task planning, various sensing systems, robot trajectory control, and intelligent welding quality controllers, corresponding system optimization design structures and system management technologies are required. From the perspective of system control development, intelligent robotic welding systems can be categorized as a complex system control problem. This problem has gained a certain academic standing in recent years of systems science research, and considerable scholars have been conducting research in this direction. Current analysis and research on such complex systems mainly focus on the combined effects of various information flows of different natures within the system, system structural design optimization, and overall system management technologies. As intelligent robotic welding control systems develop towards practical application, higher demands will be placed on their overall design and optimized management, further clarifying the importance of research in this area.
VII. Summary
In conclusion, at the current stage of welding robot technology, developing intelligent systems for welding process-related equipment is appropriate. Such systems can exist relatively independently as flexible welding product processing units (WFMC) or as sub-units of flexible manufacturing systems (FMS), offering technical flexibility. Furthermore, researching such intelligent robotic welding systems is also essential as a technological transition towards a higher goal—manufacturing highly autonomous intelligent welding robots.
