Predicting thermal cycles in duplex stainless steel

TWI Bulletin, July/August 1992

For ten years Simon Smith has been involved with numerical modelling of a variety of engineering and materials processes. He gained his Doctorate from Nottingham University in 1986, and this was followed by a postdoctoral research fellowship. At TWI Simon is a member of the Numerical Analysis Section within the Engineering Department. He has been involved in weld modelling research since joining TWI in 1989.

Robert Gunn graduated from Birmingham University with a degree in Metallurgy and Materials Science. His involvement with duplex stainless steels began at TWI in 1987 within the Stainless Steels and Corrosion Section of the Materials Department, studying the corrosion resistance of friction welds in high alloy stainless steels. In 1991 he was appointed Section Leader, but continues his studies of duplex steels, leading several projects in the area, of which one forms the basis for his doctoral thesis in collaboration with Imperial College, London.

Duplex stainless steels are susceptible to intermetallic formation during exposure to elevated temperatures, and this can reduce both mechanical and corrosion properties. Simon Smith and Robert Gunn examine how a weld modelling approach can provide a useful source of information on heat affected zone thermal cycles.

The family of ferritic-austenitic duplex stainless steels has increased in popularity over the last five to ten years, particularly in the oil and gas and petrochemical industries. This is primarily because of the attractive properties of duplex steels, namely their high strength to weight ratio (almost twice that of 316L), excellent pitting and crevice corrosion resistance, matching and often surpassing that of the AISI 316L grade, and superior chloride stress corrosion cracking resistance.

Formation of intermetallics is more prevalent in the higher alloy grades containing 25%Cr where, for example, sigma phase has been observed in the heat affected zone (HAZ) particularly when slow cooling rates have occurred ( i.e. with high heat inputs). This article describes a technique used to predict the thermal transients of a single cross-section of the HAZ, i.e. thermal cycle history, and compares the results with thermal cycles measured in bead-on-plate deposits. This work has been incorporated into a current three year project which aims to predict formation of intermetallics in the HAZ of 25%Cr duplex stainless steels.

Experimental

The test specimen is shown in Fig.1. Five metal strips were bolted together to produce a composite specimen on which 20 manual metal arc bead-on-plate passes were laid (in the sequence 1A, 1B, 1C, 1D, 2A, 2B . ..... 5D, Fig.2). All base plates were duplex stainless steel meeting UNS 532550 (selected as representative of established 25%Cr grades), whilst matching electrodes were used. No preheating was undertaken and a maximum interpass of 150°C was specified. Thermocouple measurements (both harpooned, and pre-positioned within drilled holes) were used to measure the interpass conditions accurately and to measure the heat affected zone (HAZ) thermal transients. Arc energies and travel speeds were nominally 2.0 kJ/mm and 100 mm/min respectively. Modelling was of a two dimensional cross section, taken from a bead positioned at the centre of the block width (set 3, Fig.2). Specifically weld beads 3B and 3C were of most interest as the thermocouples were positioned directly under these beads.

The thermocouples therefore recorded the highest peak temperatures and the temperature transients between 1200 and 800°C and between 800 and 500°C for these weld beads. Specific welding parameters for these beads are given below:

Model details

Mesh

A two dimensional finite element mesh was used. The model was of a single cross-section transverse to the welding direction. A complete mesh plot is given in Fig.3. As shown, the model is of one half of the complete specimen, with the reinforcement (representing weld bead 3B or 3C) on the line of symmetry. Strictly, the specimen will not be symmetrical when bead set three is deposited, since the reinforcements due to bead sets one and two will exist and those due to bead sets four and five are yet to be deposited. This slight degree of asymmetry was ignored.

The reinforcement shape was taken from a weld bead created using similar welding conditions. Although the actual specimen was composed of five plate sections bolted together, these were assumed to be in good thermal contact and were modelled as a single unit.

Thermocouple positions have been located by their distances (on the plane of symmetry) from the fusion boundary. The mesh was created so that nodes on this boundary were at equivalent positions (nodes 11-18, Fig.4). The predicted boundary was slightly smaller than the real weld, so that these nodes were approximately 0.05mm from the real thermocouple positions.

The meshes were generated using the pre- and post-processor MENTAT. ^[1] The simulations were performed using the MARC ^[2] finite element code.

Boundary conditions and heat input

A two dimensional finite element mesh was used. The model was of a single cross-section transverse to the welding direction. A complete mesh plot is given in Fig.3. As shown, the model is of one half of the complete specimen, with the reinforcement (representing weld bead 3B or 3C) on the line of symmetry. Strictly, the specimen will not be symmetrical when bead set three is deposited, since the reinforcements due to bead sets one and two will exist and those due to bead sets four and five are yet to be deposited. This slight degree of asymmetry was ignored.

The reinforcement shape was taken from a weld bead created using similar welding conditions. Although the actual specimen was composed of five plate sections bolted together, these were assumed to be in good thermal contact and were modelled as a single unit.

Thermocouple positions have been located by their distances (on the plane of symmetry) from the fusion boundary. The mesh was created so that nodes on this boundary were at equivalent positions (nodes 11-18, Fig.4). The predicted boundary was slightly smaller than the real weld, so that these nodes were approximately 0.05mm from the real thermocouple positions.

The meshes were generated using the pre- and post-processor MENTAT. ^[1] The simulations were performed using the MARC ^[2] finite element code.

Boundary conditions and heat input

Non-linear boundary conditions were assumed and the heat input was time-varying. Heat losses were due to both convection and radiation. A convective heat loss coefficient of l0 W/mK was used and radiative heat losses were with an emissivity of 0.25. ^[3]

The heat model was a volumetric thermal energy flux into the weld reinforcement. This is a simplification of the real heat flow behaviours in the actual weld where convective flows occur in the weld bead. As behaviour inside the weld bead was not of primary interest, these convection currents were not modelled and heat flow was assumed to be by conduction only.

This meant that the heat input model was achieved on a trial and error basis. Repeat simulations were performed to determine the heat input conditions which most accurately predicted the fusion boundary. There were, however, constraints placed on the allowable heat input model. The total heat input per mm weld length was assumed to equal the arc energy times the welding efficiency (80% for manual metal arc welding). In addition, the heat input was assumed to be during the period that the weld pool passes the specific location of the model. The assumed weld pool length was 20mm and the welding speed was 1.8 mm/sec. The period of heat input was therefore 11.0sec.

Material properties

Material properties for the analysis were taken from two sources. Thermal conductivity for the duplex stainless steel was assumed to be similar to that of austenitic stainless steel, which meant that the temperature dependence of this property could be taken from weld modelling literature ( Fig.5). ^[3]

Also shown on this figure are values of thermal conductivity for duplex stainless steels. ^[4] There is good agreement between the values for the duplex and those for the stainless steel up to around 400°C, which gives confidence that these values are suitable for the modelling. At higher temperatures, the alloy content of austenitic and ferritic steel alloys does not have a significant effect upon the conductivity. ^[5] The curve shown in Fig.5 is therefore suitable for the duplex stainless steel.

The specific heat capacity data came from a number of sources. ^[3,4,6] The assumed specific heat capacity versus temperature is shown in Fig.6. At lower temperatures all sources give similar values. For higher temperatures (above 1000°C), however, the duplex values are much higher than those for stainless steel. The values for duplex have therefore been used.

A duplex stainless steel manufacturer also supplied latent heats for the liquid to δ-ferrite phase transformation (1440-1340°C) and the δ to γ-austenite phase transformation (1340-1330°C). These values were respectively 414.0 kJ/kg and 27.15 kJ/kg. Since these transformations are at very similar temperatures, the modelling assumed a single phase transformation occurred between 1440 and 1330°C and released 441.0 kJ/kg of latent heat over that temperature range.

Solution procedure

MARC finite element transient heat transfer solutions are produced with a backward difference time stepping method. Accurate solutions can therefore be achieved with relatively long time steps. ^[7] The solution accuracy is controlled by the specified maximum allowable temperature change per time step. The current analysis was performed using a value of 50°C.

Results

At a number of positions (between 0.4 and 5.5mm from the fusion boundary, see Fig.4) on the plane of symmetry, the transient temperature response is given in Fig.10. These have been converted into peak temperatures and cooling times and are plotted in Fig.11 and 12. Both figures show results from thermocouple measurements taken from the experimental weld. Peak temperatures are slightly over predicted ( Fig.12). A consistent difference between prediction and experiment exists over the full range of peak temperatures and positions. Cooling times for the range 800 to 500°C were predicted with good accuracy when compared with experimental results ( Fig.12).

Some scatter exists in the thermocouple results, but the trend of both experimental and predicted cooling times is that they are almost independent of the distance from the fusion boundary. Expressed as an average for the positions shown in Fig.12, the predicted value is within 0.1% of the experimental data. Prediction of the cooling time between 1200 and 800°C is less accurate, although, considering the spread of experimental data, the agreement is reasonable.

Summary

Agreement between computed and measured thermal transients was good. This confirms that the assumed boundary conditions and material properties were representative of the real weld. The heat input model was empirical and was created by use of the known welding conditions and of dimensions taken from a sample weld (fusion boundary, reinforcement shape and crater length). Further research is necessary to determine, for a range of welding conditions, the effect of HAZ thermal cycles upon the microstructure of high alloy duplex stainless steels. TWI is presently active in this area. This work has shown that a weld modelling approach can provide a useful source of information on HAZ thermal cycles.

References