In the automotive industry, the pursuit of lightweight and high-performance components has led to the widespread adoption of aluminum-silicon cast alloys, particularly for critical parts like cylinder heads. Among these, A356 aluminum alloy is favored due to its excellent castability, thermal conductivity, and high strength-to-density ratio. However, as a component operating under cyclic mechanical stresses and thermal shocks, the cylinder head must exhibit superior mechanical properties, especially fatigue resistance. From a safety and reliability standpoint, understanding the fatigue behavior of cast aluminum alloys has become increasingly crucial. My research focuses on investigating how casting defects inherent in the manufacturing process impact the fatigue performance of A356 aluminum alloy cylinder heads under conditions simulating actual service environments.
Casting defects are unavoidable in aluminum alloy castings, and their nature and severity depend heavily on the casting method and process parameters. Common defects include oxide inclusions, shrinkage porosity, gas pores, and others. These imperfections act as stress concentrators, initiating cracks under cyclic loading and significantly reducing fatigue life. Compared to defect-free material, the presence of casting defects can degrade fatigue life by an order of magnitude. While factors like grain size (secondary dendrite arm spacing), alloy composition, modification effectiveness, and heat treatment also influence fatigue behavior, casting defects are often the dominant factor controlling fatigue strength and life in real-world components. In this study, I examine the fatigue performance of A356 alloy specimens extracted directly from production cylinder heads, aiming to correlate the types of casting defects observed with fatigue failure mechanisms and quantify their detrimental effects.
The cylinder heads used in this investigation were manufactured using a gravity permanent mold casting process with resin sand cores. This method is typical in the industry but can introduce specific defect populations. All test specimens were machined from the cylinder head body after a T6 heat treatment (solution heat treatment followed by artificial aging), ensuring they represent the actual material condition in service. Due to geometric constraints of the cylinder head, the fatigue specimens had a non-standard, relatively small size, necessitating the design of a custom fixture for mounting on the testing machine. The fixture consisted of split collets to facilitate specimen installation and ensure proper alignment during testing.
The material is A356 aluminum alloy. Its chemical composition, determined via spectrometry, is summarized in Table 1.
| Element | Si | Mg | Cu | Ti | Fe | Mn | Ni | Zn | Al |
|---|---|---|---|---|---|---|---|---|---|
| Content (wt.%) | 7.1 | 0.29 | 0.06 | 0.16 | 0.08 | 0.05 | 0.02 | 0.06 | Bal. |
Fatigue testing was conducted on a servo-hydraulic testing machine under fully reversed axial loading (tension-compression) with a sinusoidal waveform. The stress ratio R was defined as:
$$R = \frac{\sigma_{\text{min}}}{\sigma_{\text{max}}} = -1$$
where $\sigma_{\text{max}}$ and $\sigma_{\text{min}}$ are the maximum and minimum stresses in the cycle, respectively. The stress amplitude $\sigma_a$ is:
$$\sigma_a = \frac{\sigma_{\text{max}} – \sigma_{\text{min}}}{2} = \frac{\Delta \sigma}{2}$$
Tests were performed at a frequency of 20 Hz. To simulate engine operating conditions, specimens were heated and maintained at a temperature of $150 \pm 3^\circ \text{C}$ using a environmental chamber. The primary objective was to determine the fatigue strength of the material at $10^4$ cycles (a region often relevant for high-load, low-cycle fatigue in components). Two statistical methods were employed to analyze the fatigue data: the staircase (up-and-down) method and S-N curve fitting.
Prior to fatigue testing, tensile tests were performed on identical specimens at $150^\circ \text{C}$ to establish the yield strength and ultimate tensile strength (UTS), which helps set appropriate stress levels for fatigue tests. The tensile test was conducted at a strain rate of 0.3 mm/min. The engineering stress-strain curve is shown in Figure 1, from which a yield strength ($\sigma_{0.2}$) of approximately 230 MPa and an UTS of 327 MPa were determined. This indicated that fatigue tests would be conducted at stress amplitudes in the high-stress, low-cycle fatigue regime, near but below the yield point.
The casting process itself is a critical source of defects. Modern foundries use automated systems to improve consistency, but defects remain a challenge.
Such automated pouring lines aim to minimize turbulence and oxide formation, yet the complex geometry of a cylinder head can still lead to defect formation in certain regions.
For the staircase method, an initial stress amplitude step of 10 MPa was used. If a specimen survived $10^4$ cycles, the stress for the next specimen was increased by one step; if it failed before $10^4$ cycles, the stress was decreased by one step. After the first run-out (survival to $10^4$ cycles), the step size was reduced to 5 MPa to increase the resolution near the fatigue strength. The test sequence and results are plotted in Figure 2, where circles represent failures and squares represent run-outs.
The data from the staircase test were processed according to standard practice (e.g., GMN7152). Let $S_0$ be the lowest stress level in the analysis, $d$ the step size, $i$ an index for stress levels (starting at 0 for the lowest), and $N_i$ the number of survived specimens at stress level $i$. The fatigue strength (FS) at $10^4$ cycles is estimated by:
$$\text{FS} = S_0 + d \left( \frac{\sum i N_i}{N} \pm \frac{1}{2} \right)$$
where $N$ is the total number of survived specimens, and the sign depends on whether survivors or failures are fewer. The standard deviation (SD) is estimated as:
$$\text{SD} = 1.62 \times d \times (C + 0.029)$$
with the convergence factor $C$ given by:
$$C = \frac{N \sum i^2 N_i – (\sum i N_i)^2}{N^2}$$
The calculations based on the survivor data (the less frequent outcome in this test) are summarized in Table 2.
| Stress Amplitude, $\sigma_a$ (MPa) | Index $i$ | Number of Survivors $N_i$ | $i \cdot N_i$ | $i^2 \cdot N_i$ |
|---|---|---|---|---|
| 210 | 2 | 3 | 6 | 12 |
| 205 | 1 | 5 | 5 | 5 |
| 200 | 0 | 6 | 0 | 0 |
| Totals: $N=14$ | $\sum i N_i = 11$ | $\sum i^2 N_i = 17$ | ||
With $S_0 = 200$ MPa, $d = 5$ MPa, and using the survivor data ($b = +1/2$), the computed mean fatigue strength is $\text{FS} = 206.4$ MPa. The convergence factor $C = 0.60$ (>$0.30$, valid), and the standard deviation $\text{SD} = 5.09$ MPa. The mean fatigue strength value, often taken as $\text{FS} – 3 \times \text{SD}$ for a conservative estimate, is:
$$\text{MV}_{\text{staircase}} = 206.4 – 3 \times 5.09 = 191.1 \text{ MPa}$$
This represents the estimated fatigue strength at $10^4$ cycles and 150°C with a high probability of survival.
In parallel, constant amplitude fatigue tests were conducted over a broader stress range to generate data for an S-N curve (stress-life curve). The results are plotted in Figure 3, showing the number of cycles to failure $N_f$ versus the stress amplitude $\sigma_a$. The data exhibit considerable scatter, which is typical for cast materials due to the random size and location of casting defects.
The S-N data were fitted using a power-law (Basquin) relationship of the form:
$$\frac{\Delta \sigma}{2} = \sigma_a = A (2N_f)^{-b}$$
where $A$ and $b$ are material constants. A least-squares regression on the logarithmic data yields:
$$\sigma_a = 439.5 \times (2N_f)^{-0.07466}$$
The coefficient of determination $R^2$ for this fit is 0.6016, indicating moderate correlation due to scatter. The standard error of the estimate (standard deviation) for the fit is 7.148 MPa. Substituting $N_f = 10^4$ cycles into the equation gives a predicted stress amplitude of:
$$\sigma_a (10^4) = 439.5 \times (2 \times 10^4)^{-0.07466} \approx 209.8 \text{ MPa}$$
Using a similar $-3\sigma$ approach, the mean fatigue strength from the S-N curve is:
$$\text{MV}_{\text{S-N}} = 209.8 – 3 \times 7.148 \approx 188.4 \text{ MPa}$$
This value is remarkably close to the 191.1 MPa obtained from the staircase method, confirming the consistency and reliability of the experimental results despite the inherent scatter caused by casting defects. The slightly larger standard deviation from the S-N analysis reflects the broader stress range considered and the influence of extreme defect populations on fatigue life.
To understand the source of scatter and the mechanisms of fatigue failure, I conducted detailed fractographic analysis on all broken specimens using optical microscopy (OM) and scanning electron microscopy (SEM). Selected fracture surfaces were also examined by energy-dispersive X-ray spectroscopy (EDS) to identify the chemical nature of inclusions. A typical fatigue fracture surface consists of three distinct regions: (A) the crack initiation site(s), (B) the stable crack propagation region often showing striations or beach marks, and (C) the final fast fracture zone. In most specimens examined, a striking feature was the presence of multiple crack initiation sites on a single fracture surface. An example is shown in Figure 4, where at least four distinct initiation points are visible. This phenomenon aligns with models suggesting that at high stress amplitudes (near yield strength) and intermediate cycle counts (around $10^4$ cycles), the probability of multiple micro-cracks initiating from various micro-notches or defects is high. While the applied stress amplitude was in the elastic range, it was sufficiently high to promote crack initiation from multiple small casting defects that would remain dormant under lower, high-cycle fatigue stresses.
The primary initiators in virtually all cases were casting defects located at or near the specimen surface. The types of casting defects observed were categorized, and their frequency as crack initiators was statistically analyzed. The major defect types identified were: 1) Oxide Inclusions, 2) Shrinkage Porosity, and 3) Gas Porosity. Their characteristics and impacts are discussed below.
Oxide inclusions were the most frequently observed initiation feature, accounting for approximately 63.6% of the crack origins. These appear as dark, often irregularly shaped films or clusters within the initiation zone. SEM/EDS analysis, as shown in Figure 5, reveals that these inclusions are rich in oxygen, aluminum, silicon, and often contain other elements like magnesium, iron, and titanium. They are primarily bifilms or folded oxide layers formed due to turbulence during mold filling or pouring. The size of these oxide defects varied significantly, from a few tens of micrometers to, in one extreme case, a large patch nearly 1.5 mm in its longest dimension (Figure 6). Such large oxide agglomerates are not inherent in the melt but result from the entrainment of surface oxide films during the casting process.
Shrinkage porosity was the second most common initiator, found in about 31.8% of the fracture origins. This defect appears as interconnected micro-voids or dendritic cavities, often localized in regions that solidify last. Figure 7 shows a typical crack initiation from a shrinkage pore cluster. Shrinkage porosity creates a region of effectively reduced load-bearing area and acts as a potent stress concentrator due to its irregular, sharp morphology.
The remaining initiators (around 4.6%) were associated with gas pores, which are typically more spherical but can also serve as crack starters.
To quantitatively assess the relative severity of different casting defect types on fatigue life, I employed the Palmgren-Miner linear damage accumulation rule. For each specimen that failed, the actual cycles to failure $n_i$ at its applied stress amplitude $\sigma_{a,i}$ were recorded. The theoretical cycles to failure $N_i$ at that same stress were calculated from the S-N curve fit equation $\sigma_a = 439.5 \times (2N_f)^{-0.07466}$. The accumulated cycle ratio (RAC) for each specimen is:
$$\text{RAC} = \sum \frac{n_i}{N_i}$$
Since each specimen was tested at a single stress level, for a given specimen, $\text{RAC} = n_i / N_i$. An RAC value less than 1 indicates that the specimen failed earlier than predicted by the “ideal” S-N curve (which represents a population average including various defects), suggesting the presence of a particularly severe defect. Conversely, an RAC greater than 1 suggests better-than-average performance, possibly due to a more benign defect location or type. I calculated the RAC for all failed specimens and grouped them by the identified primary casting defect at the crack origin. The results are summarized in Table 3.
| Primary Casting Defect at Origin | Average RAC Value | Standard Deviation of RAC | Percentage of Specimens with RAC < 1 |
|---|---|---|---|
| Oxide Inclusions | 0.95 | 0.41 | 58% |
| Shrinkage Porosity | 0.62 | 0.28 | 92% |
| Gas Porosity | 1.10 | 0.35 | 40% |
The data clearly show that specimens where fatigue cracks initiated from shrinkage porosity had significantly lower RAC values (average 0.62) compared to those initiating from oxide inclusions (average 0.95). Moreover, 92% of specimens with shrinkage porosity origins had an RAC less than 1, compared to 58% for oxide inclusions. This indicates that, for a given applied stress amplitude, shrinkage porosity leads to earlier failure more consistently and severely than oxide film defects. Gas pores, in this limited dataset, appeared slightly less detrimental on average, but the sample size is small. The physical reason is that shrinkage porosity often forms interconnected networks with sharp edges, creating a more severe stress concentration factor and a larger effective flaw size than a planar oxide film of similar projected area. Furthermore, shrinkage tends to occur in bulk regions, while oxide films often lie along planes. However, the largest oxide patches can be extremely detrimental, as seen in the specimen with the 1.5 mm inclusion, which failed at a very low cycle count.
The location of the casting defect is also critical. Almost all crack initiations were at the specimen surface or within a few hundred micrometers of it. The stress intensity factor for a surface flaw is higher than for an internal flaw of the same size. The only internal crack initiation observed was from the exceptionally large oxide patch mentioned earlier. Even a large internal defect might allow for longer fatigue life compared to a smaller surface defect, according to fracture mechanics principles. The effective flaw size $a_{eff}$ for a surface defect is approximately related to the stress intensity factor range $\Delta K$ by:
$$\Delta K = Y \Delta \sigma \sqrt{\pi a_{eff}}$$
where $Y$ is a geometric factor (around 1.12 for a surface crack). Fatigue crack growth initiates when $\Delta K$ exceeds a threshold $\Delta K_{th}$. A casting defect effectively provides an initial crack size $a_0$. Therefore, the fatigue strength $\Delta \sigma_f$ for a given life can be inversely related to the square root of the initial defect size:
$$\Delta \sigma_f \propto \frac{1}{\sqrt{a_0}}$$
This explains the large scatter: a specimen with a small, favorably located casting defect will have a higher fatigue strength than one with a large defect at the surface. Shrinkage pores, often being bulkier and forming clusters, can create a larger effective $a_0$ than a thin oxide film, leading to lower fatigue strength.
My findings have direct implications for the manufacturing and quality control of aluminum cylinder heads. To improve fatigue performance, the casting process must be optimized to minimize both oxide inclusion formation and shrinkage porosity. This involves careful design of the gating and feeding system to promote laminar filling and ensure adequate feeding for solidification shrinkage. The use of advanced degassing techniques, filters, and controlled pouring systems (as hinted by the automated line image) is essential to reduce oxide and gas defects. Moreover, non-destructive evaluation methods capable of detecting critical-sized defects in high-stress regions could be employed for quality assurance.
In conclusion, this comprehensive investigation into the fatigue behavior of A356 aluminum alloy cylinder heads at 150°C and $10^4$ cycles reveals the profound and quantifiable impact of casting defects. The mean fatigue strength was determined to be approximately 190 MPa via two independent statistical methods. Fractographic analysis demonstrated that multiple crack initiations are common under these high-stress amplitude conditions. The predominant casting defects acting as crack starters were oxide inclusions and shrinkage porosity, with the latter being statistically more severe in reducing fatigue life. This underscores that not all casting defects are equal; their type, size, and location collectively dictate their fatiguability. For engineers and foundry specialists, prioritizing the reduction of shrinkage porosity through optimized casting design and solidification control may yield greater benefits for fatigue performance than focusing solely on oxide reduction. Future work could involve 3D characterization of defect populations and application of fracture mechanics models to predict fatigue life based on quantified defect distributions, further strengthening the link between casting quality and component reliability.
The presence of casting defects remains the principal challenge in achieving the full fatigue potential of cast aluminum alloys. As the demand for lighter and more efficient engines grows, a deeper understanding and tighter control over these defects will be paramount. This study contributes to that understanding by providing empirical data and analysis directly from production components, highlighting that in the quest for performance, the battle against casting defects is a critical frontier.