The integration of automated assembly technologies, particularly robotic riveting, represents a significant advancement in the manufacturing of large, complex structural components such as monolithic, machined舱段 (sections) for aerospace applications. These structures, often fabricated from high-strength casting parts, are subjected to dynamic loads during overland transport prior to final assembly and service. The integrity of their riveted joints under these long-duration, cyclic loading conditions is a critical design and reliability concern. This investigation focuses on a comparative analysis of the fatigue life and failure modes between traditional manual hammer riveting and modern robotic hammer riveting processes, specifically for joints connecting aluminum casting parts. The study aims to provide a statistical and mechanistic foundation for qualifying automated riveting processes for use in critical structural assemblies.
The substrate materials for all test coupons were aluminum alloys, selected to represent typical casting parts used in airframe structures. The primary material was ZL205A cast aluminum alloy, joined to 2A14 aluminum alloy plates. The use of ZL205A is particularly relevant as it is a common choice for high-integrity casting parts requiring good strength-to-weight ratio and machinability. The nominal thickness of both plates was 5 mm. Rivets were installed using two distinct processes: conventional manual hammer riveting performed by a skilled operator, and robotic hammer riveting utilizing a pre-programmed automated system. The parameters for the robotic process (including drill geometry, spindle speed, feed rate, rivet length, pneumatic pressure, and riveting time) were optimized through a series of preceding characterization trials involving hole quality, interference fit, and static joint strength tests. This ensured the automated process was tuned to produce consistent, high-quality joints for comparison against the manual baseline.
Quasi-static single-lap shear tests were first conducted to establish the ultimate shear strength of the joint configurations. Tests were performed on a universal testing machine at a constant crosshead displacement rate of 2 mm/min. A minimum of three replicates were tested for each riveting type. The load-displacement curves exhibited the characteristic三个阶段 typical of riveted joint failure: an initial linear elastic region, a non-linear region marking the onset of plastic deformation in the rivet and surrounding material, culminating in a peak load, followed by a sudden drop indicating rivet shear failure and loss of connection.
| Riveting Process | Specimen 1 Max Load (N) | Specimen 2 Max Load (N) | Specimen 3 Max Load (N) | Average Max Shear Load, \( F_{ult} \) (N) |
|---|---|---|---|---|
| Manual Hammer | 4622.4 | 4584.0 | 4766.4 | 4657.6 |
| Robotic Hammer | 4690.1 | 4611.7 | 4720.8 | 4674.2 |
The results, summarized in the table above, show that both processes yielded joints with very similar average ultimate shear strengths (\( F_{ult} \approx 4666 N \)), indicating that under static loading, both methods can achieve comparable load-bearing capacity. This average ultimate load served as the baseline for defining the fatigue load levels.
Fatigue tests were conducted on a servo-hydraulic fatigue testing machine. A sinusoidal load waveform was applied with a stress ratio of \( R = 0.1 \) (\( R = F_{min}/F_{max} \)) and a frequency of 20 Hz, selected to simulate vibrational frequencies encountered during transport. Based on the ultimate shear strength \( F_{ult} \), four maximum load levels were chosen for testing: 70%, 65%, 60%, and 55% of \( F_{ult} \). For each load level and riveting process, three to four specimens were tested to failure. The number of cycles to failure \( N_f \) was recorded for each specimen. The resulting fatigue life data for both joint types is presented below.
| Load Level (% of \(F_{ult}\)) | Max Load, \(F_{max}\) (N) | Fatigue Life, \(N_f\) (cycles) | Mean Life (cycles) | |||
|---|---|---|---|---|---|---|
| 70% | 3260.3 | Manual: 26,125 | 42,208 | 222,375 | 74,367 | 112,472 |
| Robotic: 59,817 | 96,171 | 181,427 | – | |||
| 65% | 3027.4 | Manual: 101,982 | 120,027 | 166,683 | 183,066 | 177,534 |
| Robotic: 152,289 | 177,990 | 202,323 | – | |||
| 60% | 2794.6 | Manual: 231,475 | 275,598 | 310,594 | – | 358,172 |
| Robotic: 174,430 | 367,163 | 532,923 | – | |||
| 55% | 2561.7 | Manual: 740,396 | 662,747 | 688,448 | 595,488 | 648,894 |
| Robotic: 458,922 | 702,331 | 785,429 | – | |||
A primary observation is the significant scatter in the fatigue life data for manual riveting, particularly at the 70% and 55% load levels where single specimens exhibited lives far outside the cluster of other data points (e.g., 222,375 and 740,396 cycles). This scatter is attributed to the inherent variability in manual riveting quality, which is highly dependent on operator skill, strike force, and sequence. In contrast, the robotic riveting data shows considerably less scatter at each load level, demonstrating the superior consistency and repeatability of the automated process. This consistency is paramount for reliable life prediction in critical casting parts assemblies.
The fatigue data was modeled using the Basquin equation, which relates the applied stress amplitude (or maximum load \(F_{max}\) in load-controlled tests) to the number of cycles to failure \(N_f\):
$$ F_{max} = a (N_f)^b $$
where \(a\) and \(b\) are material/process constants. The fitted curves for the two data sets reveal a critical finding: at higher load levels (70%, 65%), the fatigue lives of manual and robotic rivets are comparable. However, as the load level decreases (60%, 55%), the robotic riveted joints consistently demonstrate a longer fatigue life. The Basquin equation fits the robotic data with high correlation (\(R^2 > 0.94\)), confirming it as a good predictive model for this process. The fit for the manual data is poorer (\(R^2 < 0.78\)), even after removing obvious outliers, underscoring the statistical unreliability introduced by process variability.
To enable reliability-based design, the fatigue life data for the more consistent robotic riveting process was analyzed using a two-parameter Weibull distribution. The Weibull cumulative distribution function for failure is given by:
$$ F(t) = 1 – R(t) = 1 – \exp\left[-\left(\frac{t}{\alpha}\right)^\beta\right] $$
where \(t\) is the fatigue life (cycles), \(R(t)\) is the reliability, \(\alpha\) is the scale parameter (characteristic life), and \(\beta\) is the shape parameter (indicating dispersion). Linearizing the function allows estimation of \(\alpha\) and \(\beta\) from probability plots. The parameters for each load level were estimated as follows:
| Load Level | \(F_{max}\) (N) | Scale Parameter, \(\alpha\) | Shape Parameter, \(\beta\) |
|---|---|---|---|
| 70% | 3260.3 | 1.326e5 | 1.71 |
| 65% | 3027.4 | 1.904e5 | 2.10 |
| 60% | 2794.6 | 4.256e5 | 1.69 |
| 55% | 2561.7 | 6.489e5 | 3.85 |
The fatigue life for a desired reliability level \(R\) can then be calculated using:
$$ N_{R_x} = \alpha \left[-\ln(R_x)\right]^{1/\beta} $$
Plotting \(F_{max}\) against \(\log(N_f)\) for different reliability levels (e.g., R=90%, 50%, 10%) generates a family of S-N curves. The curve for R=90% is particularly valuable as a conservative design guideline for high-criticality applications involving casting parts, providing a fatigue life that 90% of the joints are expected to exceed.
Macroscopic examination of failed specimens revealed two primary failure modes. Failure Mode I, which was by far the most common for both processes and all load levels, involved fracture through the rivet shank within the shear plane of the overlapping plates. This mode mirrors the failure location observed in quasi-static shear tests. Failure Mode II, observed only occasionally in manual riveting specimens, involved fracture initiating at the junction of the rivet’s manufactured head and the shank. The prevalence of Failure Mode I suggests that the primary fatigue-critical location is the shank under combined shear and bending stresses induced by lap joint eccentricity.
Scanning Electron Microscopy (SEM) was employed to investigate the micro-mechanisms of fatigue failure. Fracture surfaces from Failure Mode I typically showed a classic fatigue fracture morphology. The fracture origin was located near the surface of the rivet shank, often at a point of high stress concentration or pre-existing micro-flaw. The region surrounding the origin exhibited fine, radial “clam shell” marks indicating the initial, slow crack growth stage. Beyond this, a larger area showed fatigue striations, which are microscopic markers of crack advance per load cycle. The final fast fracture region displayed a mixture of quasi-cleavage facets and micro-void coalescence (dimples), indicative of a mixed ductile-brittle final failure. The presence of these distinct zones confirms a fatigue-driven failure mechanism originating from stress concentrators.
Fracture surfaces from the less common Failure Mode II showed some differences. While a fatigue origin was still identifiable, the initial crack growth region sometimes exhibited features related to crystallographic cleavage along grain boundaries. The crack propagation area contained a higher density of equiaxed dimples, suggesting a greater influence of tensile stresses normal to the fracture plane during the later stages of crack growth. Second-phase particles were observed within these dimples. The overall appearance suggested a fatigue process where crack initiation was strongly influenced by the geometry and local microstructure at the rivet head fillet, a location potentially more sensitive to improper manual rivet set-up or hammer impact.
The contrast in fatigue performance and data scatter between the two riveting processes can be fundamentally linked to joint quality and consistency. Manual hammer riveting is susceptible to variations in rivet driving force, alignment, and final formed head geometry. This can lead to non-uniform interference fit, undesirable rivet shank bending, micro-cracking in the rivet or the adjacent casting parts material, and increased local stress concentrations. These defects act as premature initiation sites for fatigue cracks. At high loads, the dominant macro-stress state may overshadow these micro-defects, leading to similar lives. However, at lower loads, the fatigue process is dominated by the crack initiation phase. Manual joints, with their inherent flaws, initiate cracks almost immediately, leading to shorter total life. The superior and consistent interference fit and head formation achieved by robotic riveting minimize these initiation sites. Consequently, in the low-load, high-cycle regime, robotic joints spend a significantly greater number of cycles in the crack initiation phase, resulting in a longer total fatigue life—a crucial advantage for durability.
The statistical reliability analysis using the Weibull distribution is only meaningful for the robotic process due to its inherent consistency. The ability to generate reliability-based S-N curves provides a powerful tool for probabilistic design and risk assessment. When designing structures incorporating large casting parts joined by rivets, engineers can select a design curve corresponding to an acceptable probability of failure, moving beyond deterministic safe-life approaches.
In conclusion, this comprehensive study on riveted joints pertinent to monolithic casting parts assemblies yields the following key findings:
1. While both manual and robotic hammer riveting can produce joints with equivalent static shear strength, their fatigue performance diverges significantly, especially in the high-cycle fatigue regime. Robotic riveting offers superior and more predictable fatigue life at lower load levels, which is representative of long-duration transport vibration environments.
2. The fatigue life data for manual riveting exhibits high statistical scatter due to process variability dependent on human skill. Robotic riveting produces highly repeatable results, enabling accurate fatigue life prediction through models like the Basquin equation and facilitating rigorous reliability analysis via the Weibull distribution.
3. The dominant failure mode for riveted joints in simulated transport loading is fatigue fracture through the rivet shank (Failure Mode I). Occasional failures at the rivet head (Failure Mode II) in manual specimens are linked to process-induced stress concentrations. Fractographic analysis confirms classical fatigue mechanisms with initiation at stress concentrators.
4. The implementation of automated robotic riveting for assembling critical casting parts structures is strongly justified from a durability and reliability perspective. It mitigates the quality variability of manual processes, leading to longer and more predictable service life under dynamic loads, thereby enhancing the structural integrity and safety of the final product. This research provides essential quantitative data and mechanistic understanding to support the qualification and widespread adoption of automated assembly for advanced casting parts in aerospace and other high-performance industries.