Adv. Manuf. (2013) 1:112–122
`DOI 10.1007/s40436-013-0014-5
`
`Framework on robotic percussive riveting
`for aircraft assembly automation
`
`Feng-Feng Xi • Lin Yu • Xiao-Wei Tu
`
`Received: 23 December 2012 / Accepted: 23 February 2013 / Published online: 6 April 2013
`Ó Shanghai University and Springer-Verlag Berlin Heidelberg 2013
`
`Abstract Presented in this paper is a framework for the
`implementation of a robotic percussive riveting system, a
`new robot application for aircraft assembly. It is shown here
`that a successful robot application to the automation of a
`process requires in-depth research of the process and the
`interaction with the robot. For this purpose, a process plan-
`ning-driven approach is proposed to guide a robot applica-
`tion research. A typical process planning will involve a list of
`key considerations including: process sequence, process
`parameters, process tooling, and process control. Through
`this list, a number of key research issues are identified for
`robotic percussive riveting, such as rivet pattern planning,
`rivet time determination, rivet tooling design and rivet
`insertion control. The detailed research on these issues has
`effectively created know-how for the successful implemen-
`tation of our robotic percussive riveting system.
`Keywords Aircraft assembly Assembly automation
`Robotic riveting Percussive riveting Process planning
`
`F.-F. Xi (&) L. Yu
`Department of Aerospace Engineering, Ryerson University,
`Toronto, ON, Canada
`e-mail: fengxi@ryerson.ca
`
`F.-F. Xi
`Robotic and Automation Center, Shanghai University,
`Shanghai 200072, People’s Republic of China
`
`X.-W. Tu
`Department of Automation and Instruments, Shanghai
`University, Shanghai 200072, People’s Republic of China
`
`123
`
`1 Introduction
`
`two primary joining
`Riveting and welding represent
`methods for the assembly of structural components that
`require strong joint strength. Compared with welding
`mainly a fusion method, riveting a mechanical method
`generates no thermal deformation, hence widely used for
`joining high thermal conductive materials such as alumi-
`num sheet metals used in aircraft assembly [1]. There are
`hundred thousands of rivets in a regional aircraft and
`millions in a large continental aircraft. Overall, the oper-
`ation of aircraft assembly is divided into three stages:
`subcomponent assembly, component assembly, and line
`assembly. The subcomponent assembly is the first step to
`construct the base components for four major sections,
`namely,
`fuselage, wing, cockpit and empennage. The
`component assembly is the middle step to join the sub-
`components to form an individual major section. The line
`assembly is the last step to assemble a whole aircraft by
`connecting the four major sections together.
`The current riveting process in aerospace manufacturing
`entails a mix of manual riveting, semi-automated riveting,
`and automated riveting. The semi-automated and auto-
`mated riveting machines are widely used in North America
`and Europe, but only limited to component assembly, such
`as wing skin panels and fuselage skin panels. Subcompo-
`nent assembly and line assembly are still conducted man-
`ually. The labor incurred producing these subassemblies/
`assemblies amounts to as much as fifty percent of the total
`cost. Manual riveting operations are tedious, repetitious,
`prone to error, and likely causing health and ergonomic
`problems [1].
`In principle, there are two riveting methods, the first
`called squeezing (or one-shot) riveting, where a large
`upsetting force is applied to deform a rivet instantly. This
`
`RoboticVisionTech EX2012
`ABB v. RoboticVisionTech
`IPR2023-01426
`
`

`

`Aircraft assembly automation
`
`113
`
`method requires a large riveter operating under high pres-
`sure beyond the yield strength of aluminum rivets in a
`range over 500 lb force. As shown in Fig. 1a, this type of
`riveter is made of either a hydraulic cylinder or an elec-
`tromagnetic piston, very heavy, bulky and usually needing
`a lift-assisted device if used for manual operation. The
`automated and semi-automated riveting machines employ
`this type of riveter; hence they are gigantic and only lim-
`ited to riveting large, simple and relatively flat components.
`The second method is called percussive (or hammering)
`riveting, where a small
`impulsive force is applied to
`deform a rivet accumulatively by a series of hits. As shown
`in Fig. 1b, this method uses a rivet gun in size of a regular
`hand-held power tool, very compact and light, operating
`under much lower pressure in a range less than 100 psi,
`very safe and energy efficient. Manual riveting employs
`this principle.
`Research on robotic riveting has been mainly centering
`on squeezing riveting that utilizes heavy-duty industrial
`robots of large size ([100 kg payload). In the automotive
`industry, squeezing robotic riveting systems have been
`fully developed and commercialized for joining metal
`parts. This technology is called the robotic self-piercing
`riveting, in which a C-frame tooling, as shown in Fig. 1c,
`and is designed to have a squeezing riveter mounted on one
`end as a punch and the other end serving as a hitting base
`[2]. This system has been widely used for automotive
`chassis assembly. The application of robotic technology in
`aerospace manufacturing has been significantly slower than
`that in automotive manufacturing [3]. Though not com-
`mercially available, squeezing robotic riveting systems
`have been researched in the past by Boeing [4] and recently
`by EADS in Germany affiliated with AirBus [5]. In addi-
`tion, a robotic system has been implemented at Bombardier
`in Montreal that uses two giant Kuka robots to hold large
`
`(a)
`
`(b)
`
`(c)
`
`Fig. 1 Various rivet tools a Squeezing riveter: length [2400, weight
`[50 lbs, b Percussive rivet gun: length \1000 weight \5 lbs, c C-
`frame rivet tooling
`
`panels that are riveted on a C-frame squeezing riveting
`machine [6].
`Though adhesives are used to bond composites, riveting
`remains as a primary method for joining composite panels
`where there is requirement for strong joint strength and
`prevention of laminate de-bonding. Automated squeezing
`riveting systems have been developed by AirBus and
`Boeing for riveting composite panels of fuselages and
`wings. As composites are being introduced to replace steels
`for fabrication of automotive structural parts, robotic riv-
`eting will likely take over welding as a primary joining
`method for the future of automotive industry [7].
`By comparison, robotic percussive riveting is much
`more compact. Not only a much smaller riveting gun is
`used but also a light/medium-payload industrial robot of
`small size (\50 kg payload) can be applied. The overall
`system compactness offers a great advantage that a robotic
`percussive riveting system is able to access tight and
`awkward areas that a squeezing robotic riveting system is
`not able to. This advantage is referred to as good tool
`accessibility. In this paper, a framework of research is
`presented to show that the successful implementation of a
`robot application requires in-depth process research perti-
`nent to the application.
`
`2 Framework overview
`
`Prior to describing the framework, a brief summary of
`industrial robots is provided as background information.
`Generally speaking, industrial robots are general-purpose
`load-carrying motion devices. Common robot specifica-
`tions include: payload, workspace (reach), speed/accelera-
`tion and accuracy (repeatability). Industrial robots are
`normally classified according to payload as: light-payload
`(\15 kg), medium payload (\50 kg),
`large payload
`(\300 kg), and heavy duty (300–1,500 kg). While robot
`workspace is proportional to payload, the rest of specifi-
`cations are disproportional.
`The main structure of industrial robots is serial, though
`parallel robots are being used in some applications. Serial
`robots are designed to mimic human arms, often called
`articulated robot arms. Though various serial robots were
`studied, only two types have been adopted as the main
`stream of industrial robots. The first is selective compliant
`articulated robot arm (SCARA), a 4-axis robot that imitates
`the movement of a human arm when sit, as shown in
`Fig. 2a. The primary use of SCARA robots is for electronic
`component assembly. The second is programmable uni-
`versal manipulation arm (PUMA), a 6-axis robot
`that
`emulates the movement of a human arm when stand, as
`shown in Fig. 2b. The majority of industrial robots fall
`under this category. Both types of robot are designed with a
`
`123
`
`

`

`114
`
`F.-F. Xi et al.
`
`decoupling feature between positioning (first 3 axes) and
`orienting (remaining axes). The positioning of SCARA
`follows a cylindrical coordinate system, while that of
`PUMA follows a spherical coordinate system.
`The initial application of industrial robots was concen-
`trated on pick-and-place applications with grippers as the
`main tooling. Since then, they have been explored for various
`manufacturing applications both contact and non-contact. A
`good example of non-contact robot application is robotic
`welding, widely practiced in the automotive industry, where
`a weld gun is used not in contact with the workpiece. A good
`example of contact robot application is robotic polishing,
`widely adopted in the automotive and aerospace industry,
`where a polishing tool is used in contact with the workpiece.
`The robot companies who sell robots for these applications
`usually offer add-on modules in the general robot motion
`planning software. This indicates that the successful imple-
`mentation of a robot application requires understanding of
`the process itself and interaction with the robot.
`For this reason, we have developed a process planning-
`driven approach for the development of our robotic per-
`cussive riveting system, a new application for aircraft
`assembly automation. Figure 3 shows
`the developed
`robotic percussive riveting system. It includes a 6-DOF
`industrial robot that replaces the first worker for holding/
`moving a percussive rivet gun; a 5-axis computer numer-
`ical control (CNC) gantry system that replaces the second
`worker for holding/moving a bucking bar. The entire riv-
`eting process
`is
`automated through synchronization
`between the robot and gantry. Furthermore, the choice of a
`gantry system instead of a second robot allows it to serve as
`a jig for mounting sheet metals.
`Given that the goal of process planning is to generate a
`plan in order to successfully produce a product [8], this
`planning contains a number of key considerations: process
`sequence, process parameters, process tooling, and process
`control. As outlined in Table 1, these key considerations
`
`(a)
`
`(b)
`
`Fig. 2 Two common industrial robots. a 4-axis SCARA robot. b 6-
`axis PUMA robot
`
`123
`
`Fig. 3 Developed robotic percussive riveting system [13]
`
`Table 1 Process-planning driven research
`
`Process planning
`
`Required research
`
`Process sequence
`
`Process parameters
`
`Process tooling
`
`Process control
`
`Riveting pattern planning
`
`Riveting time determination
`
`Rivet tooling design
`
`Rivet insertion control
`
`are used to guide our research for the purpose of imple-
`menting our
`robotic percussive riveting system. The
`research methods are described as follows.
`
`3 Process planning-driven approach
`
`3.1 Process sequence
`
`Process sequence deals with the steps of a process, which is
`process specific. Commercial robot planning programs do
`not provide this feature. Hence, there is a need to study the
`riveting process. Riveting requires drilling a hole first and
`then inserting a rivet for fastening. Two riveting processes
`are exercised in practice, simultaneous and sequential. The
`first one is to drill and rivet together on a rivet spot, which
`demands a large tooling to combine a drill and a rivet gun.
`This process is typically applied on large automated riv-
`eting machines where tool accessibility is not of concern
`for the assembly of large and flat panels. The second pro-
`cess is to drill a series of holes first at the required rivet
`spots and then switch to a rivet gun for riveting. This
`process keeps the tooling compact and light, with good tool
`accessibility for
`tight and awkward areas, hence is
`employed for manual operation. Since our system is
`developed to replace manual operation, the second process
`is considered here.
`
`

`

`Aircraft assembly automation
`
`115
`
`Rivet Joints
`
`Lap Joint
`
`Butt Joint
`
`Single Cover
`
`Double Cover
`
`Single row
`
`Double row
`
`Multiple row
`
`Fig. 4 Taxonomy of rivet patterns
`
`Fig. 5 Riveted joints a Lap joint, b Butt joint (single cover), c Joint
`strength
`
`In riveting, the required rivet spots are determined by
`rivet patterns in light of industry standards. For automation,
`there are two ways to obtain the coordinates of these spots.
`The first one is to extract from CAD models, as modern
`aircraft components are designed using CAD. The second
`way is to compute these spots directly according to rivet
`patterns. While our development can accommodate both,
`only the second case is presented here, as the first case is
`straightforward.
`Figure 4 provides a taxonomy of rivet patterns. As
`depicted in Fig. 5, lap joints are formed by overlapping two
`pieces of sheet metal, which is asymmetric causing the
`secondary bending [9]. Butt joints, on the other hand, are
`created by aligning two pieces of sheet metal to maintain
`symmetry. Both joints can be laid out in single, double and
`multiple rows as depicted in Fig. 6. Rivet patterns can be of
`chain type with rows lined up forming a grid or of zigzag
`type with rows offset up or down.
`According to the aerospace standards [10], rivet size
`should be determined first with respect to the thickness of
`sheet metals, followed by rivet spacing. Symbolically, rivet
`size can be expressed as
`½
`Š ¼ f1 t1; t2
`ð
`Þ
`where d and l are the diameter and length of a rivet,
`respectively; t1 and t2 are the thickness of two pieces of
`sheet metal joining together. The computed rivet size must
`be rounded up to match with that specified by the stan-
`dards, such as Air-force Navy (AN), Military Standard
`(MS), and National Aircraft Standard (NAS) [10].
`Rivet spacing is a function of the rivet diameter and the
`riveting pattern, expressed as
`
`s; m½ Š ¼ f2 d; nð
`Þ
`where s and m are the spacing between adjacent rivets and
`the margin on all sides, respectively; and n denotes the
`number of rows. For given width of sheet metals, the
`number of rivets is determined from s and m. Aircraft rivets
`include solid rivets (requiring bucking bar) and cheery
`
`d; l
`
`ð1Þ
`
`ð2Þ
`
`Fig. 6 Rivet rows a Single row, b Double row, c Multiple row
`
`rivets (no bucking bars). Clecos are used as temporary
`fasteners. In this study, only solid rivets are considered.
`A rivet planning software package has been developed
`that can compute all the rivet spots based on the above-
`mentioned information. Figure 7 displays two snap shots of
`the software, with the first showing the animation window
`and the second showing the rivet spot planning window.
`Furthermore, this package is being developed to include joint
`strength analysis. Joint strength is defined as the joint’s
`ability to resist against tension, shear, cleavage and peel, as
`shown in Fig. 5c. In general, the riveted joint strength is
`proportional to the number of rows, called joint efficiency
`[11]. The common failures of the riveted joints are caused by
`the in-plane force (tension-shear shown in Fig. 6), including
`breaking of the sheet at the hole section, shearing of the rivet,
`crushing of the sheet and rivet, and shearing of the hole [10].
`Joint strength is also pertinent to the type of joints, symmetric
`better than asymmetric. Joint strength analysis can be per-
`formed using standard stress analysis methods [11] or more
`advanced finite element methods [12].
`
`3.2 Process parameters
`
`As shown in Fig. 3, in percussive riveting a rivet is placed
`between a rivet gun and a bucking bar to subject
`to
`repetitive impulses from the hammer of the gun. Due to
`these impacts, the rivet is deformed plastically to join two
`pieces of sheet metal together. Upon the determination of
`riveting process sequence, both the robot holding the gun
`and the gantry holding the bucking bar can be programmed
`to follow a path specified according to a given rivet pattern
`
`123
`
`

`

`116
`
`F.-F. Xi et al.
`
`Fig. 7 Developed rivet planning software
`
`and move from spot to spot. However, this program does
`not know how much time is needed to perform riveting at
`each spot. Hence, there is a need to study process param-
`eters, which involves riveting process modeling.
`The said modeling comprises of two theories, impact
`dynamics and plasticity [13]. Impact dynamics is applied to
`model the kinetic energy generated by the percussive gun,
`and plasticity is applied to model the rivet plastic defor-
`mation caused by the impact. As shown in Fig. 8a, a per-
`cussive rivet gun is pneumatically driven and composed of
`a piston and a hammer. Under a compressed air supply, the
`piston is pushed to drive the hammer to hit the rivet. As
`illustrated in Fig. 8b, at the start point of the piston stroke,
`the air pressure on its rear end is higher than that on its
`front end, so the piston moves forward. As it moves close
`to the end of its stroke, the pressure difference on the two
`ends reverses, thereby bringing the piston back. The stroke
`cycle repeats till the air supply is turned off. The reciprocal
`of the stroke cycle time is called the triggering frequency.
`Figure 9 shows a test result of the vibration experiment
`conducted to establish an empirical relation between the
`triggering frequency and the supply air pressure.
`The key in impact dynamics modeling is to determine
`the hammer velocity hitting the rivet. First, the piston
`dynamics can be related to the air supply as, without
`consideration of friction
`mpap ¼ Ap
`
`ð3Þ
`
`where mp, ap, A and p represent the mass, acceleration,
`cross-section area
`and air pressure of
`the piston,
`respectively. If the impact between the hammer and the
`piston is assumed perfectly elastic, the total momentum
`
`
`and total energy are conserved, hence
`¼ mh vh vho
`mp vpo vp
`ð
`Þ; vp þ vpo ¼ vh þ vho;
`
`ð4Þ
`
`where vpo, vp, vho, vh represent the velocity of the piston
`and hammer before and after the impact, respectively; mh
`denotes the mass of the hammer. Note that vpo is computed
`
`123
`
`Fig. 8 Modeling of a percussive gun [13] a Schematics of a
`percussive rivet gun, b Impact modeling of a percussive rivet gun
`
`from Eq. (3) and vho is zero since the hammer is initially
`stationary; the hammer velocity can be derived from Eq.
`(4) as
`vh ¼ 2mpvpo
`mp þ mh
`
`ð5Þ
`
`The hammer velocity is the velocity hitting the rivet
`which in turns hits the bucking bar. Each hit induces a small
`rivet plastic deformation and the accumulation of a series of
`hits results in a large rivet deformation. For this reason, the
`rivet is discretized into N elements, each modeled as a
`spring-mass-damper system, as shown in Fig. 10. The spring
`forces are modeled by a bilinear stress–strain curve
`
`

`

`Aircraft assembly automation
`
`117
`
`number of hits and the time interval of hits (i.e. reciprocal
`of the triggering frequency).
`
`3.3 Process tooling
`
`Though with rivet path planned and rivet time determined,
`actual implementation still cannot be guaranteed unless the
`tool is ensured to do the job. Different riveting methods bring
`up different issues in tooling design. In the conventional
`squeezing riveting, a large static force is applied, the main
`concern being the robot rigidity to withstand the static force.
`In percussive riveting, however, a series of impulsive (rela-
`tively small) forces is applied; the main concern becomes
`robot vibration. The general guidance of robot tooling design
`states that the tool should be designed lightweight, in com-
`pact size, and with large holding force against vibrations
`[14]. In other words, the key issue is how to keep the tool
`small yet strong. Hence, there is a need to study tool design.
`Attachment of a tool to the robot end-effector will change
`the system kinematics and dynamics. With the tool mounted,
`kinematic analysis should be carried out with respect to tool
`center point (TCP) instead of the center point of the end-
`effector (usually the center point of the mounting plate for the
`industrial robot). This analysis can be readily accommodated
`by treating the tooling system as an add-on body in the multi-
`body system of the robot, as shown in Fig. 12. Therefore, the
`system dynamic equations can be given as [15]
`Mr þ Mt
`ð
`
`Þ þ Ct q; _qð
`Þ
`Þ€q þ Cr q; _qðð
`
`Þ þ Gr qð Þ þ Gt qð Þð
`¼ s JT
`t wt
`
`Þ
`
`ð7Þ
`
`where q is a vector of the robot joint displacements; Mr,
`Mt, Cr, Ct, Gr and Gt represent the matrices of robot mass,
`tooling mass, robot coupling term, tooling coupling term,
`robot gravitational term, and tooling gravitational term,
`respectively; s is a vector of joint actuation forces; Jt is the
`Jacobian of the TCP; wt is the vector of wrench (force and
`moment) at TCP. Without the tool mounted, Eq. (7) would
`
`Fig. 9 Vibration experiment on percussive rivet gun [13]
`
`containing both elastic and plastic deformation. Conse-
`quently, the dynamics comprising the hammer, rivet and
`bucking bar can be expressed by a set of 2N ? 2 first-order
`nonlinear ordinary differential equations as [13]
`_y ¼ FðyÞ
`
`ð6Þ
`
`where y is a vector representing the displacements of the N
`elements and the hammer. Note that the bucking bar is
`fixed, no displacement. Equation (6) can only be solved by
`computer. The simulation result given in Fig. 11 illustrates
`how a rivet deforms incrementally under a series of hits.
`This simulation program is being embedded into the
`rivet planning software shown in Fig. 7 for the purpose of
`computing the required rivet time at each rivet spot. At first
`the required rivet deformation is obtained by calculating
`the difference between the original rivet length and the
`thickness of two pieces of sheet metal. Then, the simulation
`program will run to determine the number of hits needed to
`produce the required rivet deformation. At
`last,
`the
`required rivet
`time can be decided by multiplying the
`
`Fig. 10 Dynamic modeling of hammer, rivet and bucking bar [13]
`
`123
`
`

`

`118
`
`F.-F. Xi et al.
`
`Table 2 Comparison of two different tooling designs
`
`Tool
`
`gmax
`
`gmin
`
`ev
`
`w1
`
`w2
`
`w3
`
`Percussive
`
`0.7265
`
`0.9646
`
`0.9436
`
`71.3292
`
`0.2161
`
`1.7424
`
`Squeezing
`
`0.3489
`
`0.8902
`
`0.7928
`
`27.4394
`
`0.2940
`
`1.4867
`
`Fig. 11 Simulation of rivet plastic deformation [13]
`
`Fig. 12 Robot system with tooling [15]
`
`be reduced by setting Mt, Ct, Gt to zero, and Jt and wt to Jr
`and wr of the robot, respectively. Apparently, the tooling
`system will affect the entire system.
`Efforts have been made to relate the system dynamics to
`the afore-mentioned three considerations of the general
`guidance, thereby creating a new theory for the tooling
`design. For this purpose, three new indices are introduced
`to evaluate the influence of the tooling system on the
`overall robot system dynamics [15]. The first one, g, is a
`kinetic energy ratio defined as
`g ¼
`
`_qTMr _q
`_qTðMr þ MtÞ _q
`which measures the energy consumption due to the robot’s
`motion relative to the total
`(robot ? tooling) kinetic
`energy. A good tooling design would yield a large value of
`g, meaning that the effect of the tooling system is almost
`negligible, thereby addressing the issue of lightweight.
`The second one, ev, is the robot vibration ratio defined as
`ev ¼ x0t
`ð9Þ
`x0
`
`ð8Þ
`
`123
`
`Fig. 13 Comparison
`b Squeezing
`
`of
`
`two
`
`tooling
`
`systems
`
`a Percussive
`
`which evaluates the influence of the tooling system on the
`robot natural frequency, where x0t and x0 are the funda-
`mental natural frequency of the system with and without
`ffiffiffiffiffiffiffiffi
`ffiffiffiffiffiffiffiffiffi
`p
`p
`the
`tooling,
`respectively.
`It
`can
`be
`shown
`that
` ev 
`gmin
`gmax
`[15]. Since g is less than 1, the tooling
`system will reduce the fundamental natural frequency,
`thereby addressing the issue of vibration.
`The third one is the dynamic manipulability ellipsoid
`(DME) that measures the acceleration capability of the tool
`tip, thereby addressing the issue of compact size. The
`acceleration of TCP can be expressed as at ¼ Jt€q þ _Jt _q.
`Assuming that the tool accelerates from rest, i.e. _q = 0, the
`TCP acceleration can be related to the joint actuation for-
`ces using Eq. (7) as [15]
`at ¼ Bts þ agt
`where Bt is the matrix associated with the joint actuation
`forces, and agt is a vector associated with the gravitational
`acceleration. Therefore,
`the DME of the TCP can be
`evaluated by solving the singular values of matrix Bt. Three
`norms can be used: (i) w1 = det(Bt), overall capability of
`TCP, where det(.) denotes matrix determinant;
`(ii)
`w2 = cond(Bt), isotropy of DME, where cond(.) denotes
`the operation of matrix condition number; (iii) w3 = rmin,
`indicating the lowest acceleration.
`In terms of the afore-mentioned three indices, the two
`tooling designs as shown in Fig. 13 are compared. It can be
`seen from Table 2 that our tooling design for robotic per-
`cussive riveting yields a better performance than the tra-
`ditional tooling design for automated squeezing riveting,
`
`ð10Þ
`
`

`

`Aircraft assembly automation
`
`119
`
`because the index values of the former are overall higher
`than those of the latter.
`feature-based method has been
`In addition, a part
`attempted to map sheet metal part features onto the tool
`approach direction (TAD). This mapping will not only help
`further tune the tooling design but also assist in deter-
`mining correct directions for the tool to have a proper
`access for drilling and riveting. Generally, aircraft sheet
`metal parts can be classified in terms of bend direction and
`curvature. As shown in Table 3, sheet metal parts can be
`flat, single curved and double curved shapes, each requir-
`ing 3, 4 and 5 degrees-of-freedom (DOFs), respectively, in
`order to position the tool normal to the surface. Though
`tool’s DOFs are identical for both small and large curva-
`tures, tool accessibility is different. The parts with large
`curvature would have tight spaces, difficult for the tool to
`access the rivet spot.
`Tool accessibility is further affected by the areas sur-
`rounding the rivet spot. For this purpose, the taxonomy of a
`typical airframe assembly is provided in Fig. 14 to look at
`overall geometric constraints. It is true to say that fuselage
`and cockpit components involve parts with small curva-
`tures, whereas wing and empennage components involve
`parts with large curvatures. Airframe is the main body of
`an aircraft, made of structural members and covered by
`skins. While bolts are used to join structural members,
`rivets are used to join skins. In other words, riveting pro-
`cess is mainly associated with aircraft skin assembly that
`may be divided into 3 steps: skin-stiffening, skin-joint, and
`skin-to-structure.
`Skin-stiffening is to provide rigid support to a skin panel
`by riveting a number of stringers at the back of the panel.
`There are different shapes of stringer including Z, L, Y, I
`and hat-shape. Tool accessibility is affected not only by the
`size and geometry of the stringer but also the spacing
`between the stringers, as shown in Fig. 15a. Skin-joint is to
`
`join skin panels; there are two main joint designs, doubler
`(symmetric, as shown in Fig. 15b) and splice (asymmetric).
`Apparently, the size and geometry of the joint will also
`affect tool accessibility. Skin-to-structure is to mount skin
`panels onto a structure member, such as wing spar (as
`shown in Fig. 15c) or fuselage longerons (not shown).
`Probably, this is the most difficult part of skin assembly as
`the tool will be confined by the structures. In other words,
`as the skin assembly steps move up, tool accessibility
`becomes worsen.
`
`3.4 Process control
`
`Process control is to study control methods for drilling and
`riveting. There are two main issues pertaining to robotic
`riveting, localization and visual servoing. Localization is to
`transfer the coordinates of the rivet spots to those in the
`robot frame. As shown in Fig. 16, a position sensor system
`is used to measure both the jig and the rivet gun. By
`measuring three tooling balls attached to the jig, the jig
`frame, denoted by sHj, can be determined using a three-
`point method [16]. Likewise, by measuring three makers
`attached to the tool, the tool frame, denoted by sHt, can also
`be determined. Then the jig frame can be expressed with
`respect to the tool frame as
`Þ1 sHJ
`
`tHj ¼ sHtð
`where xHy
`transformation
`represents a homogeneous
`matrix from a y frame to a x frame; subscript s, t and j stand
`for sensor frame, tool frame, and jig frame, respectively.
`Since the rivet spots are expressed with respect to the jig
`frame, they can be readily transferred to the tool frame
`using Eq. (11), based on which the robot can be pro-
`grammed to follow these spots.
`Visual servoing is to control the tool tip to reach each
`rivet spot precisely based on visual sensing. As mentioned
`
`ð11Þ
`
`Table 3 Mapping of sheet metal shapes to tool’s DOFs
`
`Bend direction
`
`Bend curvature (small)
`
`Bend curvature (large)
`
`Tool’s DOFs
`
`Flat
`
`Single curved
`
`Double curved
`
`N/A
`
`3 translations
`
`3 translations ? 1 rotation
`
`3 translations ? 2 rotations
`
`123
`
`

`

`120
`
`F.-F. Xi et al.
`
`Fig. 14 Taxonomy of airframe assembly
`
`Fig. 15 Skin assembly. a Skin-stiffening. b Skin-joint. c Skin-to-
`structure
`
`Fig. 16 Localization
`
`before, the riveting process adopted here is sequential, i.e.
`drilling all the holes first based on a planned path, and then
`riveting them along the same path. Though the identical
`path program is used for both, there are two sources of
`error needing advanced control for riveting. One is robot
`repeatability and another hole tolerance. Despite calibra-
`tion to reduce system errors, industrial robots still exhibit a
`repeatability problem due to various uncertainties (random
`errors). For the robot used in our riveting system, it has a
`path repeatability of 0.6 mm and a position repeatability of
`0.2 mm. In general, the hole is drilled with a tolerance of
`0.1–0.2 mm bigger than the rivet in diameter. Apparently,
`the robot repeatability would not be able to guarantee each
`time the successful insertion of a rivet from the tip of the
`rivet gun to inside the hole. Unsuccessful rivet insertion
`would cause damage to sheet metal skins or to the tooling
`system. For this reason, advanced controls are investigated.
`Our control method consists of two parts, one dealing
`with the insertion path and another dealing with the hole
`geometry. The first part
`is to carry out a continuous
`
`123
`
`measurement of tHj based on the afore-mentioned position
`sensing system; the goal is to keep track of the tool pose in
`the course of insertion. This is done based on a Kalman
`filter method [17]
`ð12Þ
`^Xkjk1 ¼ Uk1 ^Xk1jk1 þ wk1
`where ^Xk is a state vector including three variables for the
`tool position and other three for the tool orientation; Uk1
`is a state transition matrix; and wk1 is a noise vector.
`Figure 17 shows a simulation result of the relative pose
`estimation to demonstrate the effectiveness of the method.
`The second part is to determine accurately the position of
`the hole center in the jig frame, because though the holes are
`drilled according to a planned path, they will deviate due to
`various errors. This is a one-time measurement using a high-
`resolution camera mounted on the CNC gantry. Efforts have
`been devoted to improve the accuracy of the hole center
`position that is computed from a digital image. This work is
`also being applied to the tooling system self-calibration.
`
`

`

`Aircraft assembly automation
`
`121
`
`4 System implementation and concluding remarks
`
`The research results presented in this paper have been
`applied to implement our robotic percussive riveting sys-
`tem. Figure 18 shows the physical system involving three
`controllers for three subsystems, one for the robot, one for
`the tooling and one for the bucking bar gantry. All three
`controllers are integrated, with the robot controller being
`the main controller for synchronization. A complete riv-
`eting control sequence has been generated, starting from
`position the gun ? position the bucking bar ? insert
`rive ? extend the bucking bar ? rivet ? retract
`the
`gun ? retract the bucking bar ? move to the next spot.
`This control sequence has been successfully tested and
`implemented to perform percussive riveting on sheet metal
`panels and composite panels as shown in Fig. 19.
`Our experience gained through this development has
`clearly indicated that a successful robot application to the
`automation of a process requires in-depth research on the
`process and the interaction with the robot. The said
`research can be systematically carried out according to
`process planning as this plan involve a list of key consid-
`erations including: process sequence, process parameters,
`process tooling, and process control. It has been demon-
`strated that through this list, a number of key research
`
`Fig. 17 Kalman filter method—