Candidates for automation using articulated robots include machining, additive manufacturing, composite fabrication, drilling and inspection. However, when compared to more established applications such as automotive welding, the higher tolerance requirements place unique demands on automation systems.
A number of methods can be used to improve static and low-speed accuracy, including calibration of kinematic and joint stiffness parameters, joint output encoders, adaptive control that compensates for thermal expansion and feed forward control that compensates for hysteresis and external loads. Within process planning, datums can have a huge impact on process capability and simulation is often required to accurately model and optimise this.
For high-speed operation and the compensation of high-frequency disturbances, a base level of static accuracy is first required, with additional high-dynamic compensation. High-dynamic end-effectors compensate for high-frequency disturbances using inertial sensors and reaction masses. They must be located at the end-effector as inertia would prevent the entire robot structure from responding with the required accelerations.
Global measurement feedback can be provided by six degrees-of-freedom (6-DoF) measurement systems, providing high-accuracy turnkey solutions, but these are also costly and have limited capability to compensate dynamic errors. For applications where a robot is required to position or align relative to some local feature, local feedback systems can provide mature, low-cost and highly accurate operation. In this case, feedback is typically provided by a laser line sensor.
Drilling and machining
Aircraft structures are typically joined together by drilling thousands of holes through panels and installing solid rivets. These operations require positional accuracy of approximately 0.2mm with a path straightness significantly better than this. While moving along the drilling axis, the robot must resist the disturbance or drilling vibration and reaction force. Optimisation of cutting parameters is vital to minimise these disturbances. Articulated robots may undergo rotary joint reversals while interpolating a straight line, causing accuracy to deteriorate as backlash and other hysteresis effects become dominant. Process planning may, therefore, need to optimise robot programs to avoid this.
Despite challenges, state-of-the-art production facilities are able to achieve the required tolerances, such as the Airbus A320 fuselage assembly plant in Hamburg which uses seven-axis articulated robots, integrated with linear slideways, to create the orbital joints that connect fuselage sections.
Articulated robots are finally starting to make a real impact on aerospace manufacturing
Although machining is ideally carried out by dedicated CNC machine tools, which offer high stiffness and accuracy, robotic machining has some niche applications. One example is machining component interfaces during assembly (fettling) in which it may be virtually impossible for a conventional machine tool to gain access to the interface.
Other applications may not require high accuracy but flexibility and large volume are advantageous, such as trimming castings and face machining additively manufactured blanks. The vibration and reaction forces for machining operations may be higher than for drilling. It is also much more difficult to completely avoid joint reversals, making high-accuracy robotic machining extremely challenging.
Robotic accuracy starts with kinematic calibration. Essentially, this means solving for the 6-DoF errors between each of the joints in the robot’s kinematic chain, although typically some of the degrees of freedom are assumed to be negligible to simplify the maths. For example, the Denavit-Hartenberg convention reduces the six degrees of freedom for each joint to four DH parameters. Controlling a robot involves solving the inverse kinematics, which is complicated by the fact that there are many possible robot poses that could achieve a given end-effector position and orientation.
Most articulated industrial robots have encoders mounted on their drive motors, with gear reductions of at least 100. The dominant errors are hysteresis, backlash and torsional elasticity within this drivetrain. This can be bypassed by fitting high-accuracy encoders to the joint outputs and using feedback from these to control the robot through a machine tool controller. Electroimpact has taken this approach to achieve a global accuracy of approximately 0.25mm, but it greatly increases the cost of a robot. Researchers have shown that a similar accuracy can be achieved without additional hardware by including joint stiffness parameters into the kinematic model and calibration.
Adaptive robotic control
Adaptive control is a term used within control engineering to mean the varying of control parameters, such as kinematic parameters, during the control process. The term Adaptive Robotic Control has been used in a quite different sense within industry to refer to closed-loop control using an external coordinate measurement system to provide feedback on the end-effector position and orientation, which may also be called global measurement feedback or online compensation. Adaptive control could be applied to an industrial robot by adjusting the DH parameters to allow for thermal expansion, using feedback provided by thermocouples mounted on the structure of the robot.
Twenty years ago, researchers at the University of Michigan calibrated robots with joint stiffness parameters included in the kinematic model and adaptive control for thermal deformation of the links. The calibration involved measuring the coordinates of the end-effector in many poses and then solving simultaneous kinematic equations which include all of these model parameters. Following the initial calibration, temperature sensors were positioned at multiple locations on the robot structure and the coordinates of the end-effector were measured again in multiple poses while the robot was heated. These measurements were used to fit an empirical model describing the temperature dependence for the DH parameters. They achieved a positional accuracy of 0.1mm. There is a general agreement that the most significant parameters for robot accuracy, in order of importance, are:
- The DH kinematic parameters
- The additional kinematic parameter for rotation about the Y axis
- Torsional joint compliance and backlash about the Z axis, contributing up to 10% of position errors
- Thermal expansion and shape changes owing to variation in the temperature.
Hysteresis effects can happen too quickly to be fully compensated using feedback. Feed forward control anticipates the error before it occurs, allowing correction in real time. At its simplest, this can be applied to a robot that is repeating the same trajectory, with previously observed errors used to apply corrections so that successive iterations become progressively more accurate. More sophisticated feed forward control algorithms use model-based predictions to correct for machining process forces, backlash and other hysteresis effects.
Cutting disturbances, such as vibration, may be random, unpredictable and highly dynamic. The limiting factor for the frequency of dynamic control is often the inherent inertial effects of the robot. It is simply not possible for the motors to produce enough torque to accelerate the mass of the joints quickly enough. Improvement in measurement data or communication rate will, therefore, not improve accuracy.
A state-of-the-art system has been developed by Dr Zheng Wang and others at the University of Bath. This uses a hybrid approach in which high-dynamic end-effectors are used to compensate high-frequency low-amplitude errors (vibration) while a laser tracker providing feedback to the robot controller is used to compensate low-frequency high-amplitude (path following) errors.
The work carried out in Bath has found that an expensive 6-DoF tracking system is not required for the low-frequency path following compensation. A much lower-cost 3D laser tracker provides all the significant benefits, with residual errors being dominated by backlash, hysteresis and vibration – effects that any form of realtime external feedback to the robot controller are unable to correct.
The high-dynamic end-effector contains relatively small motion stages, with much faster response times, which operate independently from the robot controller. Feedback is provided by inertial measurements. Combined with the laser tracker for low-frequency compensation, this is considerably lower in cost than current 6-DoF feedback systems while offering significantly improved accuracy for highly dynamic applications such as machining.
After many years of R&D, articulated robots are finally starting to make a real impact on aerospace manufacturing. This is largely as a result of improved accuracy. Developments only just emerging from researchers suggest that they will continue to expand into additional operations.
Want the best engineering stories delivered straight to your inbox? The Professional Engineering newsletter gives you vital updates on the most cutting-edge engineering and exciting new job opportunities. To sign up, click here.
Content published by Professional Engineering does not necessarily represent the views of the Institution of Mechanical Engineers.