Structure of corrosion layers on archaeological iron artifacts from Nanhai I

Thousands of archaeological irons were excavated from the Nanhai I ship of the Southern Song Dynasty sunk in South China Sea, most of which were severely corroded and fragmented. To understand the current corrosion state and guide the restoration and protection, one piece of these iron objects was selected for analysis. Using optical microscope, scanning electron microscope, micro-laser Raman spectroscopy, infrared spectroscopy, and X-ray diffraction, it was clear that the archaeological iron was hypereutectic white iron with carbon content of about 4.3-6.69% and had experienced low melt undercooling. There were many internal cracks formed by general corrosion extending to the iron core which was a tendency of instability. At the interface between the iron and rust, there was a black dense layer enriched with chlorine and a loose outer layer in yellow. The dense layer was mainly composed of magnetite, akaganeite and maghemite, while the rust of the loose layer was about lepidocrocite, goethite, feroxyhite, maghemite and hematite. Besides, the major phases of all corrosion products were akaganeite and lepidocrocite. Numerous holes and cracks in the rust layer had no barrier ability to the outside electrolyte, so that the iron core formed many redox electrochemical sites to be general corrosion with the rust. Meanwhile, the dense rust close to the iron core was broken locally by enriched chlorine layer, which was extremely detrimental to the stability of the archaeological iron. By electrochemical impedance spectroscopy, it could be determined that the rust layers had no protective effect on the internal iron core indeed under the condition of simulating seawater, even accelerating corrosion. The mechanism of the rust growth was proposed from the shipwreck sinking to the laboratory testing to explain the entire corrosion process. Based on the concept of authenticity on the preservation, the archaeological rusted iron of the Nanhai I ship excavated from marine environment should be properly dechlorinated and necessarily stabilized, such as corrosion inhibition and reinforcement, for the rust structure and the internal iron core well-retained together.


Introduction
Nanhai I is a wooden merchant ship of the Southern Song Dynasty (1,127-1,279 A.D.) that was excavated in Guangdong Province and is now displayed in the Maritime Silk Road Museum of Guangdong [1]. Judging from the hull structure, it could be called as "Fu Ship (Fujian-style Freighter)" which belonged to a typical type in ancient China. It is the oldest, largest, and bestpreserved one among the sunken ships found so far for international trade. The wreck may be defined in the era of the mid-13th century due to the overloaded goods on board [2]. It provides numerous historical information about the Maritime Silk Road and extremely rare materials for the study of ancient overseas trade.
In ancient trade, metal raw materials and products dominate the international market, and can be used to make weapons and labor tools [3]. A huge number of daily necessities and commercial products which made of iron or steel, such as pots, nails, and long slabs, were excavated from the Nanhai I. How to stabilize them under favorable conditions against further corrosion has been a serious problem. Further corrosion of the iron objects often occurs after extraction from marine condition and exposing to a new environment that tended to transform the metal core into corrosion products. Several articles have focused on the structure and composition of the rust layers of iron or steel [4][5][6]. They believed that the distribution of different components in the rust layer had remarkable effects on the future corrosion. In marine, the long-term immersion of iron is prone to form thick rust with many different phases: oxides, oxyhydroxides, sulphides, sulphates, chlorides, etc. Besides, the prior phases of rusts are usually a complex mixed with goethite (α-FeOOH), akaganeite (β-FeOOH), lepidocrocite (γ-FeOOH), and magnetite (Fe3O4) [7]. The rust composition and distribution of archaeological irons are always more complicated than the simulated in the laboratory, especially under natural marine corrosion for more than 800 years. Determining the state of rust are beneficial to infer its corrosion mechanism and predict future corrosion for the conservationists.
This article reported the metallic phase of iron materials in the Southern Song Dynasty of China and analyzed the rust state of marine corrosion over hundreds of years to provide useful information for preventing further degradation of iron artifacts after salvation and ensuring its preservation. The iron fragments covered with rust from the Nanhai I ship were this research objects. The rust was mainly analyzed by optical stereomicroscopy, micro-Raman spectroscopy and electrochemical impedance spectroscopy combined with SEM, XRD, and FTIR to evaluate the current corrosion and try to find the corrosion causes during marine period and after salvation.

Archaeological iron objects
The iron samples were all collected from the Nanhai I ship. The initial states of these archaeological iron were multiple stacked and completely covered by rust and mud. They were placed in a pH=9 sodium hydroxide solution during transportation to archaeological laboratory.
Most of them have been completely corroded and shattered. During initial cleaning with deionized water to remove the silt and mud, the rust layers were gradually exposed. The surface appearance and the approximate dimensions shown in Fig.1.

Samples Preparation
Rust powder: The rust powder was scraped from the iron object with a razor and a brush, and then the rust was milled in an agate mortar for a uniform and fine size.
Cross-section of iron core and rust layers: The sample was cut from the edge of the archaeological iron and then dried with a dryer at room temperature to remove free water on the surface. The object was embedded in epoxy resin, and then mechanically grinding using SiC paper (grade 80-2000) under absolute ethanol. Finally, this sample was polished with diamond paste to obtain a smooth cross-section.
Working electrode: Cutting along the above cutting edge to ensure the continuity and similarity of the rust layers and the iron core. The sample was cut in a small piece of 10 mm × 10 mm. It was carefully embedded in epoxy resin to mask and protect the cut edges for electrochemical impedance test. The details were shown in Fig. 2.

Determination of rust composition
To determine the crystalline structure of the rust powder, X-ray diffraction patterns were recorded by a diffractometer (PANanalytical X'pert-3 Power, Netherlands) with Cu anode. The generator voltage was 40 kV and tube current was 40 mA. Angular scanning was performed from 5° to 80° with a scan step size of 0.013°. The rust was previously powdered and screened to a particle size of less than 125 µm.
KBr-matrix pellets were prepared to define chemical structures of rust using FTIR transmittance analysis (Bruker Tensor 27, Germany) in the ranges from 400 to 4000 cm −1 at 4 cm −1 resolution. 0.2 mg of powder sample was mixed with 200 mg of dry KBr (> 99% FTIR grade, Sigma-Aldrich), then milled in an agate mortar and pressed to be pellets. Different rust components were defined by micro-Raman spectroscopy performed on the crosssection sample using micro-Raman spectroscope (Thermo Scientific DXRxi, America) with a microscope (OLYMPUS BX51, Japan). Raman excitation was provided by a frequency doubled Nd: YAG laser operating at 532 nm with a power of about 0.2 mW and with a probe diameter of about 1 mm. All spectra were calibrated using the 519.5 cm -1 line of a silicon wafer.

Electrochemical impedance
Electrochemical impedance spectroscopy of rusted and naked archaeological iron objects were investigated using an electrochemical workstation (CorrTest TM CS-350, China). Measurements were carried out in a three-electrode cell with 1.5 g/L NaCl and 1.5 g/L Na2SO4 mixture as an electrolyte to simulate sea water. The three-electrode cell includes a saturated calomel reference electrode filled with saturated KCl solution which served as a reference electrode, a platinum auxiliary electrode with an exposure surface of 10 mm × 10 mm as a counter electrode, and the iron objects with an exposure surface of 1 cm 2 as working electrode as described above. Prior to impedance tests, the 5 electrodes were kept in the solution for 30 min to stabilize the free corrosion potential. Tests were performed at open circuit potential in an applied frequency ranged from 100 kHz to 10 mHz, and the sinusoidal perturbation signal with 10 mV amplitude was used. The obtained data were interpreted based on equivalent circuit using Zview to obtain the fitting parameters.

Results and Discussion
Cracks in the archaeological iron After grinding and polishing, the cross-section sample has shown a significant corrosion state without any corrosive agent. The cross-section had a length of 0.78 cm and a width of 0.36 cm, which was approximately a rectangular plane. General corrosion and long cracks extending from the rust layer to the iron core were clearly observed, and the state of the upper surface was the most obvious. There was a worryingly large crack in the upper left corner, which should be caused by the combined effect of general corrosion and crevice corrosion. Due to the internal stress of corrosion products [8], it would gradually expand the core and tend to peel off a small part from the metal object. The cracks in the upper right corner were densely distributed and intersected, extending in different directions to the core, which was not conducive to a stable storage of the archaeological iron. Then, this embedded sample will continue to be used to analyze the cast state of the iron core and the composition of the rust layer on the upper surface. The cross-section was silvery white. The white and long strips were primary cementite, and 7 the matrix was eutectic ledeburite. The archaeological sample was a typical hypereutectic white cast iron with high carbon content in the range of 4.3%-6.69 wt.%.

Metallographic of the archaeological iron
When the hypereutectic molten iron was cooled, the wide and coarse primary cementite was produced first. Just because it could grow freely in the liquid, it was shaped as strips or flakes. Then, the temperature of the metal liquid continued to drop, and the eutectic phase occurred to be eutectic ledeburite. Due to the local cooling, primary cementite and eutectic ledeburite were also arranged in different cooling directions [9]. Generally, two major phases of eutectics were normally encountered in white cast irons with high-carbon under different undercoolings. They could be described as ledeburite eutectic which had dendrite branches and another was plate-like eutectic.
Besides, plate-like eutectic in white cast irons became the main constituent of the microstructure with increased degree of melt undercooling (T~35-40℃), while ledeburite eutectic often appeared at lower undercooling (T~5-25℃). This transformation does not really depend on starting alloys composition (hypo-or hyper-eutectic) [9]. The ledeburite in the metallographic images of the archaeological iron section was mostly dendritic. Based on this, it could be inferred that the iron object from the Southern Song Dynasty has experienced slow cooling after casting into a mold. β-FeOOH, respectively [10]. The peak at 1026 cm −1 was due to the O-H bending in γ-FeOOH [11].

Composition of the rust powder
In addition, the characteristic peak of γ-Fe2O3 and β-FeOOH could be overlapped at 648 cm −1 , while Fe3O4 had two strong characteristic peaks between 2850-2950 cm −1 , which was also overlapped the position of β-FeOOH [12]. As for the peak of 3377 cm −1 , it was a common peak of amorphous rust [10].
Also, XRD was used to analyze the composition of rust powder, combined with data of infrared spectroscopy. XRD spectrum was compared individually by Jade software and PDF-card data. The peaks of the XRD data and the provided standard spectrum of akaganeite could clearly correspond to each other, which could be regarded as the main phase of rust powder. Meanwhile, a significant signal of pure iron appeared, proving that there were some tiny iron particles that peeled off with rust due to corrosion. The peaks intensity of other rust components were weak, but a small amount of goethite (α-FeOOH), lepidocrocite, feroxyhite (δ-FeOOH), hematite (α-Fe2O3), maghemite and magnetite could be confirmed according to the PDF information.
Ferric oxyhydroxides has four common types, where goethite was an electrochemically stable phase, whereas others were all active phase and could be transformed into relatively stable magnetite [6]. The results of FT-IR and XRD indicated that a large amount of β-FeOOH was present in the rust, which was associated with the high concentration of chloride in the South China Sea. However, in this work, α-FeOOH was only detected in the XRD with weak intensity and scarce signal in the IR spectra.
Most iron-based compounds involved in corrosion reactions were prone to transform into β-FeOOH and γ-FeOOH in marine, resulting in the little amount of α-FeOOH in rust. Generally, there was a fact that the formation of β-FeOOH need chloride ions which could stabilize the structure of its crystal [13].
Besides, β-FeOOH was the most detrimental to continuous corrosion of archaeological iron among all the ferric oxyhydroxides [14]. Therefore, the composition in the rust powder preliminarily indicated that the rust phase of these archaeological irons had not yet reached stability, and the subsequent stabilization treatment was necessarily required.
Structure of the rust layers 9 Fig. 7 Graphs of microstructure and element distribution of the rust layer A (red square for Raman analysis) Fig. 7 showed detail of the rust layer A which marked in fig. 3. The surface of the archaeological iron was mainly divided into two major rust layers: the outer layer was yellow and present a loose and porous appearance, while the inner one close to the iron core was black and dense. Large cracks penetrated the whole rust layer, even breaking into the iron core and cutting the dense black layer to be peeled off easily. The iron element was almost evenly distributed in the whole rust. The chlorine element was evenly distributed in the outer rust and formed a highlight band which coincided with the position of the dense rust layer. Micro-Raman detected eight points respectively in Fig. 8. The oxides and hydroxides of iron compounds were identified layer by layer according to the literature values [7,[15][16][17]. The crosssection of the rust layer A indicated four distinct layers based on colors: a yellow outermost layer dotted with light-yellow spots (I), a yellow layer (II), a black layer dotted with yellow spots (III), and a black innermost layer which was closely connected with the iron core (IV). Raman spectra showed that the outermost rust was mainly formed by hematite (α-Fe2O3) and maghemite (γ-Fe2O3), with characteristic peaks at 215, 278 and 390 cm −1 to α-Fe2O3 and the relatively weak peak at 1591 cm −1 to γ-Fe2O3. The light-yellow dots in the outer rust showed a strong peak at 399 cm −1 with less intense peaks at 298, 685 and 1311 cm −1 were attributed to the existence of goethite (α-FeOOH). A strongest peak closed to 251 cm -1 and a second peak in the Raman shift of 382 cm -1 was related to 11 the lepidocrocite phase (γ-FeOOH) which coexisted in the middle part of the yellow rust layer with α-FeOOH phase. The presence of feroxyhite (δ-FeOOH) was confirmed by its characteristic bands at 351 and 700 cm -1 at the junction of black and yellow rust. Akaganeite (β-FeOOH) was often regarded as the most dangerous compound in rust due to its hygroscopicity in post-excavation corrosion and easily transformation into other oxides, especially the magnetite (Fe3O4) [18]. Besides, chloride ions were often found in the crystal lattice of β-FeOOH, which could lead to complete loss of the artefact [7]. In this object, akaganeite had the bands at 137, 299 and 386 cm -1 ,while magnetite was assigned the Raman band at 347, 459, 702 and 1414 cm -1 . The two kinds of rust had obvious signals near the black dense layer, where chlorine was enriched. Area B was the rust layer on the upper right of the iron sample. There were several wide and hollow lines at the interface between rust and iron, which did not tightly wrap the internal metal to protect against external environment. The overall structure was similar to that of the layer A, except that the black rust gradually occupied the yellow outer layer. Maghemite, goethite and lepidocrocite were the dominant phases in the yellow loose rust layer, and the black dense layer where chlorine gathered was still mainly composed of akaganeite and magnetite.
Therefore, there were two obvious rust layers on the upper surface of the iron. The loose yellow rust had a tendency of exfoliation and the dense black rust had a lot of cracks, so that the inner iron core still had channels in contact with the external environment which was hard to be stable. Figs. 11 and 12 showed the Nyquist and Bode plots of naked (removing the rust) and rusted archaeological iron, respectively. For the rusted surface, the Nyquist plot was composed of a depressed capacitive semi-arc in the high frequency, and a long tail in the low frequency region representing typical Warburg impedance. The charge transfer resistance of the iron rust dominated at the low frequency region while the rust resistance dominated at the high frequency region [19].
This indicated that the electrochemical corrosion process on the rust/iron interface in NaCl/Na2SO4 solution was controlled by the diffusion process. The construction of rust layers could be described using the equivalent circuit in Fig. 13(b). After the rust surface was completely removed, the archaeological sample reacted with the simulated seawater directly. Only one double-layer capacitive semicircle was observed in the Nyquist plot and its equivalent circuit was shown in Fig.   13(a). In Figs. 9(a) and (b), RS represented the electrolyte resistance, Cr the rust capacitance, Rr the rust resistance, Cdl the double-layer capacitance, Rct the charge-transfer resistance, W the Warburg resistance or barrier diffusion impedance which was associated with the diffusion of corrosive electrolyte into the iron surface through pores in the rust layer that acted as a diffusion barrier. Due to the surface roughness and energy dispersion, a constant phase angle element (CPE) was introduced to describe the Cdl and Cr in the fitting circuits [20].
The electrochemical impedance for the circuit of Fig. 13(b) could be expressed by the following equation: where Yr and nr, Ydl and ndl were constants representing the elements Cr and Cdl, respectively, Zw was the Warburg resistance representing as follows [14]: where AW was the modulus of ZW.
The total impedance for the circuit of Fig. 13(a) could be expressed by the following equation: The equivalent circuits of Figs. 13(a) and (b) were used to fit the Nyquist and Bode plots of rusted and naked iron, respectively. Fitting data was listed in Table 1. Compared with that of the naked iron, Rct of the rusted one was much smaller, which was attributed to the reduction of the rust accelerating the cathodic reaction. The lower Rct value indicated that the corrosion products formed on the archaeological object were less compact, less continuous, and porous [19]. Rr was usually used to evaluate the corrosion resistance of the rust [13]. Therefore, it could be the rust layer had poor barrier behavior and it cannot effectively prevent the corrosion of iron core, which reached an 15 agreement with the porous structure of the rust layer mentioned above.

Mechanism of the rust growth
The rust growth and iron corrosion mechanism of this archaeological iron could be proposed from the iron surface to the outer rust layer. Since the rust growth of archaeological iron was not an independent process of a single rust phase, it was better to describe the rust growth according to environmental changes in multiple periods.
When the Nanhai I ship was hit by a shipwreck, the cargo on the ship sank into the sea. The iron products began to immerse in a salt-rich water environment and started corroding. The corrosion reactions in initial stage could occur as follows [21].
The iron objects worked as an anode region: As the ship sinks, a large amount of air would be brought in instantly and attached to the surface of the cargo in the form of bubbles. Hence, the interface between iron and sea water would be a cathodic reduction due to the oxygen dissolved in the thin water film. Meanwhile, countless electrochemical cells involving cathodic and anodic areas were started and scattered on the iron surface to form numerous corrosion sites: Then, the interface would quickly turn into normal seawater with high chloride concentration, with the consumption and bursting of tiny bubbles. The chloride ions were more readily adsorbed than oxygen in competition for surface sites and deposited on the metal with high retention promoted the formation of ferrous chloride accelerating the corrosion of iron. Besides, the accompanying hydrolysis started to create a weakly acidic environment around it [22]: Mud and sand would gradually envelop and cover the archaeological objects with the water flow. Blocking the update and circulation of dissolved oxygen and reserving the chloride ions around the iron. The high chloride concentration and local acidic conditions at the interface gave chances to the formation of FeOOH, especially akaganeite and lepidocrocite [11]: Eventually, the archaeological iron was wrapped and buried tightly under the silt in deep sea, forming an anaerobic and weakly acidic environment. Fe3O4 could be reduced and converted rapidly from γ-FeOOH and β-FeOOH [23,24]. Ferrous ions and electrons could pass through an Fe3O4 layer due to its good conductivity. Besides, Fe3O4 was prone to accumulate at the interface and form dense layer on the surface of iron under the cathodic reaction [25]. The reduction of the γ-FeOOH and β-FeOOH layer would continue to transform into the cathode area of Fe3O4 layer. Meanwhile, the chloride ions would gradually cause local break of the oxide/oxyhydroxide film on the iron objects, especially at rough and uneven surface [26], so the corrosion process of the archaeological iron was slow but unstoppable in the marine for more than 800 years: After being excavated out of seawater, the archaeological iron was immersed in a weakly alkaline solution and transported to the laboratory for research. For preservation and display, this iron object was placed at room temperature to gradually remove internal moisture. Since the oxygen concentration and humidity changed significantly on the rust surface during drying stage, it was prone to form γ-FeOOH again and α-FeOOH. At this stage, γ-FeOOH and α-FeOOH could be generated by oxidizing the precipitation of Fe(OH)2 derived from the outer rust, and only a fraction of FeOOH was generally adsorbed on the surface of outermost layer. The structure of newly generated FeOOH was loose and did not cover the objects tightly resulting in the formation of pores in yellow rust. Hence, when the archaeological iron was removed from dechlorination solution or rinsed with distilled water, it was easily detached. After the immersing in alkaline solution, FeOOH in the dehydration stage was more likely to form maghemite and hematite [18]:

Conclusions
Archaeological iron artifacts excavated from the Nanhai I ancient ship were corroded and fragmented on a large scale. The metallographic structure of them belonged to hypereutectic white cast iron with carbon content of 4.3-6.69 wt.% and experienced low melt undercooling. There were many cracks in the iron core caused by general corrosion, which was the direct cause of the fragmentation.
Further, the upper rust of the archaeological iron was most corroded which could be distinguished in two layers. The outer layer was loose yellow rust, mainly composed of α-Fe2O3, γ-Fe2O3, α-FeOOH, γ-FeOOH, and δ-FeOOH. The dense black rust layer close to the iron core was gathered β-FeOOH and Fe3O4 with full of threatening chlorine which could break this dense layer.
Meanwhile, there were also many cracks in the rust layers extending to the iron surface which resulting in a very low resistance against the seawater indicated that the rust was a poor barrier of the internal metal. Both were the important reasons for general corrosion of the archaeological iron.
There was a reasonable mechanism proposed to explain the growth and transformation of each rust layer and the reason why the rust layer cracked and lost its protective effect, which combined corrosion conditions and reactions from the initial stage being submerged in seawater to the state before laboratory protection.
For the better conservation of these archaeological irons, the rust should be rigorously dechlorinated and stabilized with proper corrosion inhibition and reinforcement. Moreover, the preservation status should be monitored in daily routine.