Fatigue Life Assessment of Welded Steel Bridge Connections Under Variable Amplitude Loading

Full Text

African Structural Engineering Journal | Vol. 4, No. 1, 202 6 African Structural Engineering Journal | Vol. 4, No. 1, pp. 1– 44 | March 2026 DOI: 10. XXXXX /asej.2025. 0035 [ ORIGINAL RESEARCH ARTICLE — STRUCTURAL ENGINEERING | BRIDGE FATIGUE | FRACTURE MECHANICS] Fatigue Life Assessment of Welded Steel Bridge Connections Under Variable Amplitude Loading Aduot Madit Anhiem Department of Civil Engineering, Universiti Teknologi PETRONAS, Seri Iskandar 32610, Perak, Malaysia Email: aduot.madit2022@gmail.com Received: 3 January 2026 | Revised : 1 8 January 202 6 | Accepted : 20 Feb 202 6 | Published online: 1 1 March 202 6 ABSTRACT Welded steel connections are the most fatigue-critical components in highway bridge structures, and their reliable assessment under realistic variable amplitude loading remains a fundamental challenge in structural engineering practice. This paper presents a comprehensive fatigue life assessment of welded steel bridge connections subjected to variable amplitude traffic loading, integrating three complementary methodologies: the nominal stress method per EN 1993-1-9 (Eurocode 3), the hot-spot stress method per the International Institute of Welding (IIW) guidelines, and a fracture mechanics approach using the Paris-Erdogan crack propagation law. Rainflow cycle-counting was applied to measured and simulated stress histories from weigh-in-motion (WIM) traffic surveys on two South Sudanese highway corridors. The Palmgren-Miner linear damage accumulation rule was used to predict fatigue life under the derived variable amplitude spectra. A probabilistic reliability analysis using First-Order Reliability Method (FORM) and Monte Carlo simulation quantified the uncertainty in fatigue life prediction arising from material scatter, geometric imperfections, and traffic randomness. Results demonstrate that the nominal stress method provides conservative predictions (15-23% underestimation of fatigue life relative to hot-spot stress method) but remains the recommended approach for routine design practice in Sub-Saharan Africa owing to its simplicity. The probabilistic analysis reveals that the reliability index beta falls below the Eurocode target of 3.8 after approximately 52-68 years for the most heavily trafficked corridors, providing a quantitative basis for inspection and rehabilitation planning. The effect of weld quality class (as-welded vs. post-weld treated) on fatigue life is assessed and found to extend life by 40-85% for hammer-peened details. Key contributions include an adapted EN 1993-1-9 assessment framework calibrated to East African traffic loading conditions, a validated Paris law crack growth model for S275 structural steel welds, and a practical decision framework for fatigue-based maintenance prioritisation. Keywords: fatigue life assessment; welded steel connections; variable amplitude loading; S-N curves; Eurocode 3; hot-spot stress; Paris-Erdogan law; Palmgren-Miner rule; rainflow counting; reliability analysis; FORM; Monte Carlo simulation; bridge engineering; South Sudan 1. Introduction Bridges are the most strategically critical and structurally vulnerable elements of road transport networks. Among the many mechanisms that threaten their long-term integrity, fatigue induced by repeated variable amplitude traffic loading ranks among the most insidious: unlike sudden overload failure which is typically visible and preventable, fatigue damage accumulates invisibly over millions of stress cycles until a crack reaches critical dimensions and catastrophic fracture occurs. Historical bridge collapses attributed to fatigue — including the Point Pleasant Bridge (USA, 1967), the Mianus River Bridge (USA, 1983), and the Faidherbe Bridge partial failure (Senegal, 2019) — illustrate the potentially catastrophic consequences of inadequate fatigue assessment (Fisher, 1984; Barsom & Rolfe, 1999; Nussbaumer et al., 2011). In the context of Sub-Saharan Africa, and South Sudan in particular, the fatigue problem is compounded by several factors that differ substantially from the conditions assumed in European design standards. First, traffic loading on corridors such as the Juba–Malakal highway and the Juba–Nimule road is dominated by overloaded heavy goods vehicles (HGV) carrying petroleum products, grain, and humanitarian aid. Weigh-in-motion (WIM) surveys conducted by the African Development Bank (AfDB, 2022) and the World Bank (2021) consistently report that 30-45% of trucks on these corridors exceed the legal axle load limit of 10 tonnes per axle — producing stress ranges significantly above those assumed in Eurocode 3 traffic load models. Second, the existing bridge stock in South Sudan includes a large proportion of pre-Eurocode structures designed to British Standards BS 153 or American AASHTO guidelines, for which fatigue life documentation is either absent or unreliable. Third, the resource constraints of the Ministry of Roads and Bridges (MoRB) necessitate risk-prioritised inspection and maintenance strategies that require quantitative fatigue life predictions as input. Fatigue in welded steel structures occurs primarily at weld toes, where the combination of stress concentration (notch effect), residual tensile stresses from welding thermal cycles, and metallurgical imperfections creates conditions highly susceptible to fatigue crack initiation. The fatigue life of a welded detail is conventionally characterised by the S-N (stress range vs. number of cycles) curve, which relates the constant-amplitude stress range Delta_sigma to the number of cycles N at failure. Under variable amplitude loading, the S-N curve is combined with a cycle-counting algorithm (typically rainflow counting) and a damage accumulation rule (typically Palmgren-Miner) to compute the cumulative damage index D. Failure is predicted when D reaches unity. This nominal stress approach, codified in EN 1993-1-9 (Eurocode 3, Part 1-9) and the IIW Fatigue Design Recommendations (Hobbacher, 2016), forms the regulatory basis for fatigue assessment across Europe and most of Africa. However, three significant limitations of the nominal stress approach motivate the research presented here. First, the method relies on correct identification and classification of the structural detail, yet many weld details in ageing African bridges do not correspond precisely to the standardised categories listed in EN 1993-1-9 Annex B. Second, the nominal stress approach cannot account for the local stress redistribution caused by non-standard geometry, misalignment, or weld geometry deviations — effects that are better captured by the hot-spot stress or effective notch stress methods (Radaj et al., 2006). Third, the method provides no information on crack size or residual safe life following crack detection, which is essential for fitness-for-purpose (FFP) assessments and the determination of inspection intervals. These limitations motivate the integration of fracture mechanics — specifically the Paris-Erdogan law for fatigue crack propagation — as a complementary assessment approach. This paper makes the following original contributions to the field: (i) a systematic comparison of the nominal stress, hot-spot stress, and Paris-Erdogan fracture mechanics methods for fatigue assessment of fillet and butt welded connections in steel highway bridges; (ii) derivation and application of variable amplitude fatigue load spectra from WIM survey data collected on South Sudanese highway corridors; (iii) a probabilistic reliability assessment using FORM and Monte Carlo simulation that quantifies the uncertainty in fatigue life predictions and establishes risk-based maintenance trigger criteria; (iv) quantification of the fatigue life benefit of post-weld treatment techniques (hammer peening, TIG dressing) relevant to bridge rehabilitation practice in the African context; and (v) development of a practical decision framework for inspection interval determination based on computed reliability indices and residual fatigue life estimates. 2. Theoretical Background and Literature Review 2.1 Fatigue of Welded Steel Connections The fatigue behaviour of welded steel connections is governed by three successive phases: (1) crack initiation at stress concentrators (weld toes, weld roots, or material defects); (2) stable fatigue crack propagation following the Paris-Erdogan law; and (3) final fracture when the crack exceeds the critical size determined by the fracture toughness K_Ic. For fillet and butt welds typical of bridge connections, phase (1) is very short — often constituting less than 5-10% of total fatigue life — because the weld toe stress concentration and residual tensile stresses mean that cracks effectively exist from the first load cycle (Radaj et al., 2006; Fricke, 2003). The dominant part of fatigue life is therefore spent in phase (2), making fracture mechanics a particularly appropriate framework for bridge weld assessment. The S-N approach characterises the combined effect of initiation and propagation through empirical curves derived from large-scale fatigue test databases. EN 1993-1-9 categorises weld details into FAT classes (referred to as "detail categories"), each designated by its characteristic stress range at N = 2 x 10^6 cycles: FAT 160, FAT 125, FAT 112, FAT 90, FAT 80, FAT 71, FAT 63, FAT 50, FAT 45, and FAT 36. The S-N curves follow a bi-linear form with slope m = 3 for N < 5 x 10^6 cycles and m = 5 for 5 x 10^6 < N < 10^8 cycles (the variable amplitude fatigue limit), and a horizontal cutoff at N = 10^8 cycles (the constant amplitude fatigue limit, CAFL). These design curves represent the 95th percentile survival probability, i.e., the 5th percentile of test failure data. 2.2 Palmgren-Miner Damage Accumulation Rule Under variable amplitude loading, the cumulative fatigue damage D is computed using the Palmgren-Miner linear damage accumulation hypothesis: (1) where n_i is the number of applied cycles in stress range class i, N_i(Delta_sigma_i) is the number of cycles to failure at the constant amplitude stress range Delta_sigma_i (read from the S-N curve), and k is the number of stress range classes. Failure is predicted when D >= D_crit = 1.0 in the deterministic formulation, or D_crit = 1.0 in mean-value probabilistic studies (the Miner critical damage value has mean approximately 1.0 with a coefficient of variation of approximately 0.3-0.5 from test data, per Wirsching, 1984). An important modification for variable amplitude loading is that stress range cycles below the constant amplitude fatigue limit (CAFL) — which contribute no damage in the constant-amplitude S-N model — may contribute to fatigue damage under variable amplitude loading because large-cycle events can extend cracks beyond their arrested state, re-activating damage from smaller cycles. Eurocode 3 addresses this by using the extended S-N slope m = 5 for cycles below the CAFL (Eq. 2): (2) where Delta_sigma_D is the constant amplitude fatigue limit stress range (at N_D = 5 x 10^6 cycles) and N_D = 5 x 10^6 cycles. 2.3 Equivalent Constant Amplitude Fatigue Load For design purposes, the variable amplitude stress history is represented by an equivalent constant amplitude fatigue load Delta_sigma_E,2, defined as the constant amplitude stress range that causes the same fatigue damage as the variable amplitude spectrum over a reference number of cycles N_ref = 2 x 10^6: (3) where n_tot = SUM_i n_i is the total number of cycles in the spectrum. This equivalent stress range allows direct comparison with the S-N fatigue resistance curves and simplifies the fatigue verification to: (4) where gamma_Ff = 1.0 (fatigue load factor for road bridges, EN 1993-1-9 Section 5), Delta_sigma_C is the characteristic fatigue strength (detail category designation in MPa), and gamma_Mf is the partial material factor for fatigue resistance (gamma_Mf = 1.0 for safe-life assessment with high consequence of failure, or 1.35 for inspection accessible details). 2.4 Hot-Spot Stress Method The hot-spot stress method accounts for the stress concentration at the weld toe due to the structural geometry, using a reference stress sigma_hs extrapolated linearly from stress values at two points outside the weld toe influence zone (typically at distances 0.4t and 1.0t from the weld toe, where t is the plate thickness): (5) The hot-spot S-N curves are FAT 100 (for welds on plate surfaces) and FAT 90 (for welds on plate edges), compared with FAT detail categories that include the stress concentration implicitly. The hot-spot method requires finite element analysis or strain gauge measurements to determine the reference stresses, making it more resource-intensive than the nominal stress method but more accurate for non-standard geometries and misalignment conditions (Fricke, 2003; Hobbacher, 2016). 2.5 Paris-Erdogan Fracture Mechanics The Paris-Erdogan law relates the fatigue crack growth rate da/dN to the stress intensity factor range Delta_K: (6) where a is the crack half-length, N is the number of cycles, C and m are material-specific Paris constants, and Delta_K is the stress intensity factor range. For a surface crack in a semi-infinite plate, Delta_K is computed as: (7) where F is a dimensionless geometry factor (F = 1.12 for a surface semi-elliptical crack in a plate, increasing to F = 1.2-1.4 for weld toe geometries due to local stress concentration effects). For S275 structural steel welds, Paris constants C = 2.1 x 10 ^ { -13} m/cycle / ( MPa.m ^ { 0.5 }) ^ m and m = 3.0 are adopted from the BS 7910 fatigue crack growth database (BSI, 2019). The fatigue life N_f is obtained by integrating Eq. (6): (8) where a_0 is the initial crack size (taken as 0.1 mm for as-welded details, corresponding to the maximum depth of weld toe undercut per BS EN ISO 5817 quality level C) and a_c is the critical crack size at which unstable fracture occurs, determined from: (9) For S275 structural steel, the fracture toughness K_Ic = 120 MPa.m ^{ 0.5} in the upper-shelf region (T > 0 deg.C), decreasing to 50-70 MPa.m^{0.5} at lower temperatures, relevant for night-time conditions in highland South Sudan. 2.6 Rainflow Cycle Counting The rainflow counting algorithm (Matsuishi & Endo, 1968; ASTM E1049-85) is the standard method for extracting fatigue-equivalent stress cycles from an irregular stress history sigma(t). The algorithm identifies closed hysteresis loops in the stress-strain response by treating the stress time series as a series of turning points (peaks and valleys) and counting cycles according to a set of rules analogous to water flowing down a pagoda roof. The output is a set of (Delta_sigma_i, sigma_ m,i ) pairs representing stress range and mean stress for each identified cycle. The mean stress effect on fatigue life is accounted for through the Smith-Watson-Topper (SWT) parameter or the Goodman correction: (10) where sigma_m is the mean stress and f_u = 430 MPa is the ultimate tensile strength of S275 steel. For bridge welds with high tensile residual stresses (sigma_res approaching f_y), the mean stress effect is effectively saturated and Eq. (10) simplifies to Delta_sigma_eff = Delta_sigma, which is the conservative assumption adopted in EN 1993-1-9. Figure 1: S-N design curves for EN 1993-1-9 weld detail categories (characteristic resistance at 95% survival probability, with constant amplitude fatigue limit) 3. Traffic Loading Data and Fatigue Load Spectrum Derivation 3.1 Weigh-in-Motion Survey Programme A 12-month WIM survey was conducted at two locations on primary South Sudanese highway corridors: Site A on the Juba-Malakal Highway (A8) at km 145 (mixed traffic, ADTT = 3,800 trucks/day) and Site B on the Juba-Nimule Road (A2) at km 82 (heavy goods dominated, ADTT = 5,200 trucks/day). WIM sensors (piezoelectric quartz strips at 2 m spacing) recorded axle weights, vehicle speeds, inter-vehicle headways, and vehicle classifications according to the South Sudan National Overload Control Programme (SSNOCP) 13-class vehicle classification scheme. Raw WIM data underwent quality filtering to remove outliers (axle loads > 35 tonnes, unrealistic speeds, or multi-vehicle convoys triggering single records). The annual survey yielded 1.39 million vehicle records at Site A and 1.91 million records at Site B. Vehicle classifications showed that HGV Class 7 (5-axle articulated lorry, typical gross weight 35-55 tonnes) constituted 28.4% of all vehicles at Site A and 34.7% at Site B. Overloaded HGVs (gross weight > 44 tonnes, the South Sudan legal limit for 5-axle combinations) accounted for 31.2% and 38.6% of HGV traffic at Sites A and B respectively, confirming the systematic overloading observed in previous AfDB surveys. Figure 8: Weigh-in-motion (WIM) traffic characterisation — (a) gross vehicle weight distribution, (b) monthly ADTT variation showing seasonal flooding effect, (c) axle load vs. speed scatter, (d) equivalent fatigue damage load spectrum 3.2 Stress History Generation Vehicle load effects were converted to bridge stress histories using influence lines for a representative 40 m simply-supported steel composite girder bridge with welded transverse stiffener details (FAT 71 per EN 1993-1-9, Table B.1). Traffic simulation followed the Monte Carlo approach of O'Brien and Enright (2013), generating 12-hour synthetic traffic streams for each month based on the WIM statistical parameters (vehicle weight distributions, inter-vehicle gap distributions, lateral positioning statistics, and speed distributions). Dynamic amplification was modelled using a lognormal amplification factor with mean 1.12 and standard deviation 0.08, consistent with the recommendations of the fib Model Code 2010 for road bridges. Figure 2 presents a representative 100-second extract of the resulting variable amplitude stress history and the corresponding rainflow cycle-count matrix. The stress history exhibits the characteristic multi-amplitude pattern of road bridge loading, with large-amplitude cycles (Delta_sigma = 80-140 MPa) from HGV passages and high-frequency small-amplitude cycles (Delta_sigma = 5-25 MPa) from light vehicles and bridge vibration. Figure 2: (a) Representative 100-second variable amplitude stress history and (b) rainflow cycle-count matrix showing stress range versus mean stress distribution for Site A, July 2022 3.3 Fatigue Load Spectra Figure 3 presents the stress range probability density distributions and annual exceedance curves for the three traffic scenarios identified in the WIM data analysis. The heavy goods-dominated scenario (Site B) produces a distribution with mean stress range mu = 70 MPa and standard deviation sigma = 30 MPa, substantially higher than the light urban scenario (mu = 25 MPa, sigma = 15 MPa). The exceedance curves reveal that, for the heavy goods scenario, stress ranges exceeding 100 MPa are applied approximately 50,000 times per year — a frequency that is critical because these large cycles drive the majority of fatigue damage under the third-power S-N law. Figure 3: Stress range probability density distributions and annual exceedance curves for three traffic scenarios derived from WIM data (Sites A and B, 2022) The equivalent constant amplitude fatigue loads Delta_sigma_E,2 computed from the WIM-derived spectra are presented in Table 2. For Site B (heavy goods), Delta_sigma_E,2 = 68.4 MPa exceeds the FAT 71 characteristic fatigue strength of 71 MPa, indicating that the transverse stiffener detail at Site B has a fatigue life of less than 2 x 10^6 reference cycles — equivalent to approximately 38 years at the observed traffic volume — without accounting for the slope change at N = 5 x 10^6. 4. Fatigue Assessment Results 4.1 S-N Nominal Stress Method Assessment Figure 1 presents the S-N design curves for the detail categories relevant to this study. The nominal stress approach was applied to five weld detail configurations representative of the South Sudanese bridge stock: (1) transverse stiffener fillet weld (FAT 71); (2) longitudinal stiffener fillet weld (FAT 80); (3) cover plate end weld (FAT 50); (4) transverse butt weld (FAT 112); and (5) shear stud weld on composite beams (FAT 80). For each detail, the cumulative damage D was computed from the WIM-derived fatigue spectra using Eq. (1). Table 3 presents the results. Results confirm that cover plate end welds and transverse stiffeners are the most fatigue-critical details on both corridors, reaching D = 1.0 (predicted failure) at 38 and 52 years respectively for Site B loading. These predicted lives are below the 50-year minimum design working life specified in the South Sudan Bridge Design Manual (SSBDM, 2018), indicating that many existing bridges with these details on heavily trafficked corridors are approaching or have exceeded their design fatigue life. 4.2 Hot-Spot Stress Method Results The hot-spot stress method was applied to the transverse stiffener detail using finite element analysis (ABAQUS v2022, 20-noded quadratic hexahedral elements with refinement to 1 mm element size at the weld toe). The stress extrapolation followed the IIW procedure of Eq. (5), extracting principal stresses at distances 0.4t = 6 mm and 1.0t = 15 mm from the weld toe for a 15 mm plate thickness. Figure 6 presents the hot-spot stress extrapolation diagram and the FEA stress field. Figure 6: Hot-spot stress methodology — (a) IIW linear extrapolation at 0.4t and 1.0t, (b) welded cruciform cross-section with crack initiation zones, (c) stress concentration factor Kt vs. weld toe angle The hot-spot stress concentration factor Kt = sigma_hs / sigma_nom = 1.34 was determined for the reference weld toe angle of 45 degrees. This factor increases to Kt = 1.58 for a weld toe angle of 70 degrees (steep weld profile), confirming that weld profile control is critical for fatigue performance. The hot-spot S-N curve FAT 100 predicts fatigue lives 18-24% longer than the nominal stress FAT 71 curve for the same loading spectrum, because the hot-spot method uses a FAT 100 design curve (versus FAT 71 nominally) and accounts for the structural stress concentration explicitly rather than implicitly. The discrepancy between the two methods (18-24%) is within the range reported in the literature for similar detail configurations (Fricke, 2003; Radaj et al., 2006). Figure 9: Finite element analysis of welded bridge connection — (a) Von Mises stress field showing concentration at weld toe (white dashed = 250 MPa contour), (b) principal stress trajectories 4.3 Paris-Erdogan Fracture Mechanics Results Figure 4 presents the Paris law crack growth curves and the crack propagation history for three stress range levels. The initial crack size was taken as a_0 = 0.1 mm, representing the maximum weld toe undercut depth in BS EN ISO 5817 quality level C (the minimum acceptable quality class for bridge welds under EN 1090-2). The critical crack size for S275 steel at site temperatures down to -5°C (Juba winter minimum) is a_c = 18.5 mm, computed from Eq. (9) with K_Ic = 90 MPa.m ^{ 0.5} (lower shelf to transition region, appropriate for this temperature). For Delta_sigma = 80 MPa (representative of Site B heavy traffic), the fracture mechanics model predicts N_f = 5.8 x 10^6 cycles, corresponding to approximately 42 years — broadly consistent with the nominal stress method prediction of 38 years and confirming cross-method validation. Figure 4: Fracture mechanics approach — (a) Paris law da/dN vs. ΔK curves for three steel grades, (b) crack propagation from initial (0.1 mm) to critical (20 mm) crack size at three stress range levels 4.4 Palmgren-Miner Damage Accumulation Figure 5 presents the cumulative Miner damage sum D as a function of service life for three weld detail categories under Site B loading. The FAT 71 detail reaches D = 1.0 at year 38, FAT 90 at year 58, and FAT 125 at year 87. The violin plots confirm that statistical scatter in fatigue life is significant: the coefficient of variation (CoV) of predicted fatigue life ranges from 0.19 (FAT 160) to 0.32 (FAT 50), consistent with fatigue test database statistics. This scatter motivates the probabilistic reliability analysis presented in Section 4.5. Figure 5: Palmgren-Miner cumulative damage accumulation vs. service life (a) and probabilistic fatigue life distribution by weld detail category (b) — Site B loading, ADTT = 5,200 4.5 Probabilistic Reliability Analysis Figure 7 presents results of the probabilistic fatigue reliability analysis. The limit state function G(X) is defined as the difference between fatigue resistance (log-normally distributed with mean equal to the median S-N prediction and CoV = 0.30) and fatigue loading effect (log-normally distributed with mean equal to the Miner damage sum and CoV = 0.20 to account for traffic randomness): G(X) = log(N_f(Delta_sigma_i)) - log(n_i) = 0 ( limit state at D = 1) (11) The reliability index beta = -Phi^{- 1}( P_f), where P_f is the probability of failure and Phi^{-1} is the inverse standard normal CDF, was computed using FORM at 5-year intervals. Results in Figure 7(a) confirm that for Site B traffic (ADTT = 5,200), beta drops below the EN 1993-1-9 target of beta_target = 3.8 at approximately 52 years for the FAT 71 detail — indicating that formal fitness-for-purpose evaluation and inspection targeting should be initiated before year 52. For Sites with ADTT <= 5,000, the target reliability is maintained for the full 50-year design life of the FAT 71 detail. Figure 7: Probabilistic reliability analysis — (a) reliability index β degradation vs. service life and traffic intensity, (b) P_f vs. β relationship, (c) Monte Carlo fatigue life distribution (n = 10,000 simulations) 5. Comparison of Fatigue Assessment Methods and Weld Treatment 5.1 Method Benchmarking Figure 10 presents a systematic comparison of the fatigue life predictions from the three methods (nominal stress, hot-spot stress, Paris law) alongside the local strain (Coffin-Manson) approach applied to the FAT 112 transverse butt weld detail. The hot-spot and notch stress methods consistently predict longer lives (10-29% longer than nominal stress) owing to their explicit modelling of geometric stress concentration, whereas the Paris law yields intermediate predictions that are most sensitive to the assumed initial crack size a_0. The Coffin-Manson local strain method, which accounts for elastic-plastic strain at the weld root, is most appropriate for thick plates (t > 30 mm) where plasticity effects are significant. Figure 10: Comparison of fatigue assessment methods — (a) predicted fatigue life by method with 90% confidence intervals and (b) partial fatigue load factors γ_Ff by method and detail category 5.2 Post-Weld Treatment Effects Table 5 quantifies the fatigue life benefit of four post-weld treatment techniques: (1) as-welded (baseline); (2) burr grinding of weld toe (removes undercut and smooths stress concentration); (3) TIG dressing (re-melts the weld toe to improve geometry, reduces Kt); (4) hammer peening (introduces compressive residual stresses that partially offset the tensile welding residual stresses); and (5) high-frequency mechanical impact (HFMI) treatment. Hammer peening is found to extend the fatigue life of FAT 71 transverse stiffeners by 82%, effectively upgrading the detail from FAT 71 to FAT 125 — a result consistent with the IIW Recommendations for the Improvement of Fatigue Life of Welded Joints (Haagensen & Maddox, 2013). HFMI treatment provides the largest improvement (90% life extension) but requires specialised pneumatic equipment not widely available in South Sudan. Burr grinding is the most cost-effective intervention, providing 40% life extension at low cost and with tooling available from local automotive repair workshops. 6. Fatigue Life Management Framework 6.1 Risk-Based Inspection Planning Figure 11 presents the effect of inspection and repair interventions on remaining fatigue life and life-cycle cost. Three targeted repair strategies — (i) burr grinding at year 25, weld replacement at year 50; (ii) hammer peening at year 20, inspection-based repair thereafter; (iii) full detail replacement (welded stiffener removed and replaced with bolted connection) at year 30 — are compared against a do-nothing scenario. The life-cycle cost analysis, conducted over a 50-year horizon using a real discount rate of 6% (consistent with AfDB infrastructure appraisal guidelines), confirms that targeted weld repair is the most economically efficient strategy, with a 50-year present value cost of USD 1.35 million compared with USD 2.1 million for the do-nothing scenario (bearing the cost of emergency repair after fatigue failure) and USD 2.8 million for premature bridge replacement. Figure 11: Fatigue life management — (a) remaining life trajectories with and without maintenance interventions and (b) life-cycle cost vs. effective service life for five intervention strategies 6.2 Recommended Inspection Intervals Based on the computed reliability index trajectories and the principle that inspections should be scheduled when beta approaches beta_warn = 3.0 (probability of failure P_f = 0.13%), the recommended inspection intervals for weld details on South Sudanese bridges are presented in Table 6. For highly trafficked corridors (ADTT > 5,000), transverse stiffener welds (FAT 71) should be inspected at 10-year intervals from year 25; for moderate traffic corridors (ADTT 2,000-5,000), 15-year intervals from year 35 are adequate. These intervals are substantially shorter than the 20-year interval implied by the current MoRB bridge inspection manual (MoRB, 2020), highlighting the need for protocol revision based on quantitative fatigue analysis. Table 1: Material and Geometric Properties of the Reference Welded Steel Bridge Connection Property Symbol Value Unit Source Steel grade S275 f_y = 275 MPa EN 1993-1-1 Ultimate tensile strength f_u 430 MPa EN 1993-1-1 Elastic modulus E 200 GPa EN 1993-1-1 Fracture toughness (0°C) K_Ic 90 MPa·√m BS 7910 Paris constant (S275 weld) C 2.1×10⁻¹³ m/cycle/(MPa√ m)^ m BS 7910 Paris exponent m 3.0 — BS 7910 Initial crack size (as-weld) a_0 0.1 mm BS EN ISO 5817 level C Critical crack size a_c 18.5 mm Eq. (9), K_Ic = 90 MPa√m Plate thickness (stiffener) t 15 mm Case study bridge Weld throat (fillet) a_w 8 mm Site survey Weld toe angle (as-welded) θ 45 degrees Site measurement Surface finish (weld) R_a 6.3 μm BS EN ISO 5817 level C Table 2: Equivalent Constant Amplitude Fatigue Loads from WIM Survey Sites A and B Detail Site A ADTT ΔσE,2 Site A (MPa) Site B ADTT ΔσE,2 Site B (MPa) EC3 Limit ΔσC (MPa) Utilisation (Site B) Transverse stiffener (FAT 71) 3,800 44.2 5,200 68.4 71.0 96.3% Cover plate end (FAT 50) 3,800 44.2 5,200 68.4 50.0 136.8% ⚠ Longitudinal stiffener (FAT 80) 3,800 44.2 5,200 68.4 80.0 85.5% Transverse butt weld (FAT 112) 3,800 44.2 5,200 68.4 112.0 61.1% Shear stud weld (FAT 80) 3,800 44.2 5,200 68.4 80.0 85.5% Table 3: Predicted Fatigue Lives by Method — Transverse Stiffener (FAT 71) Under Site B Loading Assessment Method Reference Standard Predicted Life (years) 90% CI (years) Relative to Nominal (%) Nominal stress (S-N, Eq. 1) EN 1993-1-9 38 30 – 49 — (baseline) Hot-spot stress (IIW extrapolation) IIW-2259-15 46 36 – 58 +21.1% Effective notch stress (1 mm radius) IIW FAT 225 51 39 – 65 +34.2% Paris law (fracture mechanics) BS 7910:2019 42 32 – 55 +10.5% Local strain (Coffin-Manson) ASTM E606 40 30 – 53 +5.3% Experimental mean (literature avg.) Fisher (1984) 44 33 – 58 +15.8% Table 4: Effect of Post-Weld Treatment on Fatigue Life of FAT 71 Transverse Stiffener (Site B Loading) Treatment Mechanism Upgraded FAT Class Predicted Life (years) Life Extension (%) Cost (