High Accuracy Automation for Aerospace Manufacturing
Dr Jody Muelaner posted on June 17, 2019 |

Industrial robots are well-established in many industries and are commonly associated with modern advanced manufacturing systems. However, aerospace manufacturing still relies heavily on skilled manual work. A key barrier to the adoption of robots in aerospace is a lack of accuracy. With improved accuracy the aerospace industry is forecast to see a big increase in the use of flexible automation. This article looks at the technologies that can improve robot accuracy and some applications for them.

Automation in Aerospace: The Challenges

Many people are surprised to learn that aircraft assembly is still a largely manual process. Aerospace is commonly considered the pinnacle of advanced manufacturing and it is, therefore, surprising that in some ways it’s behind the traditional automotive industry. It is even more surprising when you consider the similarities between aerospace and automotive assembly.

When most people think about industrial robots, they think about automotive assembly. However, robots do almost everything now.
When most people think about industrial robots, they think about automotive assembly. However, robots do almost everything now.

Assembly can be divided into three types, according to how the form of the assembly is determined:

  • In a determinate(or determinant) assembly, only the parts determine the final shape of the assembly. This is often used for assemblies made up or ridged parts, such as an engine or gearbox. This is a bit like building Lego.
  • In a jig build, an external frame determines the final shape of the assembly. Parts which are flexible and/or have moveable interfaces are loaded into a jig which holds the emerging assembly in the desired shape. The parts are fastened in-situ to create a rigid assembly and the whole assembly is then removed from the jig.
  • In a measurement assisted assembly, parts which are flexible and/or have moveable interfaces are held relative to one another using moveable clamps and fixtures. Measurements are then used to guide the position and orientation of the parts to produce the desired shape of the assembly. The parts are fastened in-situ to create a rigid assembly and the clamps and fixtures are removed. Buildings are often constructed in this way.

Both automotive bodies and airframe structures are usually jig built. They both involve thin flexible panels with sliding interfaces. These panels are clamped into a jig and then fastened together. In automotive assembly the panels are usually joined using spot welds, while in aerospace they are usually drilled through in the jig and filled with a rivet or bolt. In aerospace assembly, it is also often necessary to shim or fettle interfaces between components. A shim is used to fill a gap, while fettling means machining away excess material. Both add weight, since fettling means that fettling allowances must be included, which aren’t always removed.

Manual drilling and filling of an aerospace assembly in a jig.
Manual drilling and filling of an aerospace assembly in a jig.

There has, in recent years, been a significant amount of automation of the drilling and filling operations in aerospace manufacturing. However, this has mostly used bespoke gantry-based machines rather than flexible automation. This is more like a very large machine tool than a welding robot. The problem with this approach is that it requires huge investment in a single design—it’s like old fashioned mass production rather than modern lean automation. A lack of flexibility causes huge problems when the production system needs to be reconfigured. This almost bankrupt Ford when Model T production ended in 1927 and will impact Airbus profits as A380 production is wound down. Airbus invested heavily in bespoke gantry automation for A380 wing panel drilling and filling.

Automated drilling and filling of Airbus A380 wing panels.(Image courtesy of Electro Impact.)
Automated drilling and filling of Airbus A380 wing panels.(Image courtesy of Electro Impact.)

There are many challenges involved in using flexible automation in aerospace. Drilling causes reaction forces and vibrations which can necessitate more rigid machinery. Fettling requires higher accuracy than current robots can achieve. Compared to other industries, aerospace assembly is complex and low volume, it involves larger numbers of unique operations to produce relatively few final products. This means that very large numbers of robot programs must be generated. A further difficulty is that because final structures are large and complex, it is necessary to perform multiple operations concurrently. This means that humans may need to work in close proximity to robots which requires greatly enhanced safety systems.

Drilling with Reduced Reaction Forces

Conventional drilling has issues with high on-axis reaction forces and levels of vibration. This makes it difficult for the relatively flexible structures of industrial robots to produce good-quality, accurate holes. Holes can also be produced by interpolating a circular path using a milling machine with a smaller diameter cutting tool. This results in lower reaction forces and lower vibration. However, heavy machine tools able to drill holes in large airframes are not flexible or reconfigurable enough for lean production systems. Conversely, industrial robots can’t interpolate circles accurately enough due to the combination of stiffness and inertial effects, as well as backlash and servo mismatch.

Orbital drilling provides a method of interpolating machined holes using flexible automation. It effectively uses a very small machine tool that has just enough movement along each axis to interpolate a hole. This orbital drilling machine is then positioned where the hole is required in the same way that a conventional drilling machine would be positioned. The machine is light enough for a robot to handle, enabling flexible and reconfigurable drilling within large assemblies.

An Orbital Drilling machine.(Image courtesy of Novator.)
An Orbital Drilling machine.(Image courtesy of Novator.)

Understanding Robot Accuracy

Many factors affect the accuracy of a robot. These include:

  • Angle encoders
    • Calibration
    • Repeatability
    • Thermal drift
    • Other drift effects
  • Dynamic drives and control
    • Drivetrain compliance
    • Backlash and other hysteresis effects
    • Controller path approximations
  • Bearings
    • Wear and play
    • Run-out
    • Elastic deformation
  • Links (bending, twisting and change in length due to loads)
    • Thermal loads
    • Gravity loads
    • Inertial loads
    • End-effector reactions

For static positioning, industrial robots can have short-term repeatability of just a few micrometers. However, repeated operations cause this repeat positioning to drift over time. The repeatability of a robot should, therefore, be expressed as a function of time. Repeatability of a few micrometers may be possible for a few minutes but these increases to hundreds of micrometers over hours and often millimeters over days. The best high-accuracy robots have long-term repeatability of less than 0.1 millimeter. Repeatability may be considered a limiting factor for accuracy.

There are many other accuracy requirements. Off-line and automated programming of robots, in which robots are programmed within a digital 3D environment, requires absolute accuracy, not simply repeatability. Accuracy requires relatively complex calibration since there are many different poses that a robot can use to reach the same end-effector position. Kinematic calibration is used in which many observations of the robot’s end-effector are used to solve for a kinematic model of the robot containing error parameters representing the six degrees-of-freedom for each link. The Denavit–Hartenberg parameters (DH parameters) apply some assumptions to simplify this to four parameters for each link.

Diagram of the classic four-parameter DH convention.
Diagram of the classic four-parameter DH convention.

Machining, and to a lesser extent drilling, require dynamic path following accuracy. This is strongly affected by several factors including:

  • Inertial forces determined by minimum path radius and velocity
  • Ability to overcome inertial forces due to stiffness and motor torque
  • Temporal resolution of controller feedback loop
  • Mechanical backlash

The best tools for measuring the accuracy of robots are laser trackers and ball bars. The laser tracker was originally invented for robot calibration but, although it has now found many other applications, it remains ideal for this purpose. A laser tracker can measure coordinates within a robot’s working volume to just a few microns. However, for measuring dynamic path following accuracy, laser trackers are less well-suited. For this, a ball bar is better-suited. The ultimate test for any production tool is, however, a coupon test. If you want to know how accurately a robot can drill a hole or machine a surface, then actually produce that feature in a test part. Such studies should create a feature multiple times and then measure the parts to determine bias and variation in the process.

In order of importance, the most significant factors affecting accuracy are:

  1. The conventional 4 DH kinematic parameters
  2. The additional kinematic parameter for rotation about the y-axis
  3. Torsional joint compliance and backlash about the z-axis
  4. Thermal expansion and shape changes due to variation in the temperature

1 and 2 are corrected by commercial high-accuracy robots achieving 0.5 mm path following accuracy—these may also correct 3. Thermal effects are not thought to be corrected in any commercially available system. Research shows potential for accuracy as low as 0.1 mm.

Adaptive Control – Parameters that Change

Adaptive control involves updating the parameters in the control model during operation. This is not to be confused with Adaptive Robotic Control, a proprietary process used by Nikon Metrology, which is actually closed loop control with fixed parameters. Implemented for an industrial robot, this would involve calibrating the stable influences and including them in the control model and correcting for transient influences. Research at the University of Michigan has used a combination of fixed parameters including the enhanced DH parameters and joint stiffness modelled as a linear spring, together with adaptive parameters for thermal expansion updated by temperature sensors. This adaptive control achieved an accuracy of 0.1 mm. Further improvements in this method may be possible by modelling additional parameters such as degrees of freedom (DoF) for joint and link compliance, and backlash compensation.

Positional Feedback from Optical Metrology Systems

One solution for achieving very high static positioning accuracy is to provide real-time feedback from a laser tracker. Laser trackers are normally used to calibrate a robot and angular encoders in each joint then provide feedback to the joints’ motors. Providing positional feedback during operation can enable holes to be drilled with an accuracy of 0.05 mm. A mature turn-key solution is available from Leica which is a hybrid of a laser tracker with photogrammetry to give 6 DoF feedback, although it is an expensive system. For dynamic path following, laser trackers are somewhat less accurate although an accuracy of 0.2 mm has been demonstrated.

Research at the University of Bath has shown that a cheaper laser tracker measuring only coordinates may be able to provide a similar level of accuracy as the robot encoder is sufficient for rotational DoF. This research has also shown that inertial effects limit the dynamic accuracy of a robot operating under closed loop control, therefore increased feedback frequency will not improve this.

Photogrammetry has also been used to provide positional feedback to robots. Conventional photogrammetry systems are somewhat limited for dynamic applications. Any photographer will understand that there is an inherent compromise between speed and resolution. If you want to capture very rapid movement then you need a fast shutter speed which means less light and more noise. This is an unavoidable optical compromise. Digital cameras also have data transfer limitations that further restrict their ability to capture high frame-rate dynamic motion at high resolution. Photogrammetry designed for accurately tracking high-speed motion, therefore use one-dimensional CCD arrays. These have greater light sensitivity and reduced data requirements. Three linear arrays can determine the coordinates of a target. The speed and accuracy is, however, still lower than a laser tracker with positioning accuracy of 0.2 mm. Nikon Metrology has used this technology in their Adaptive Robotic Control.

High Frequency End-Effector Actuation

Since the limiting factors for dynamic control of a robot are the inherent inertial effects, additional actuation is required to improve on this. In other words, it’s not possible for the robot to respond fast enough to the feedback it receives. Therefore, improving the accuracy or frequency of feedback won’t help. High-dynamic end-effectors have been developed at the University of Bath that can respond to feedback independently from the robot controller. These small motion stages can achieve very rapid response times. This can be deployed with a hybrid approach in which:

  • A laser tracker provides feedback to the robot controller, compensating for low-frequency high-amplitude (path following) errors.
  • High-dynamic end-effectors compensate high-frequency low-amplitude errors (vibration)

Conclusions

Advances such as orbital drilling and high-accuracy robots are enabling flexible and lean automated production systems for aircraft and spacecraft. Work is also being carried out to address the need for humans to operate in close proximity to robots and the challenges of programming many unique operations for high complex products. Further advances using adaptive control and hybrid feedback will lead to further increases in the usefulness of robots over the coming years.


Recommended For You