The longitudinal forces generated from the Friction Stir Welding process are substantial. An understanding of these forces is critical to proper tool design. This study describes a technique to measure the longitudinal force distribution on a friction stir weld pin tool. Total longitudinal forces were recorded on a dynamometer while welding 6061 aluminum with non-threaded pins that varied in length and diameter. A model was developed that characterizes pin force as a function of pin length and diameter. Results suggest that force generally increases with pin length, while forces remain relatively constant with pin diameter. Unexpected force variation was found at large pin lengths, which yielded several possible models of the force distribution. All of the modeled force distributions proved to be non-uniform and increase linearly with pin length, which produces a pin force that increases with the square of the pin length.

Quantitative understanding of frictional phenomena between the tool and the workpiece is essential for accurate modeling of the Friction Stir Welding (FSW) process. Two methods of measuring the tool-workpiece interface are proposed that allow frictional measurements to be made under extreme conditions. The first method uses a cylindrically curved surface in contact with a flat plate. The ranges of temperature, velocity, and normal force used in this method are 100–600°C, 0.38–2.0 m/s (75–400) surface feet per minute (SFM)), and 450–2700 N (100–600 lbf), respectively. Data are gathered at different parameter level combinations to provide enough data to create an empirical model representing the data. Two friction modes with distinct characteristics are observed. One mode, Coulomb-Amonton's friction, has frictional force proportional to normal force, while the other mode, plastic shear deformation friction, has frictional force independent of normal force. A linear statistical model has been developed to characterize each of the frictional modes for the polycrystalline cubic boron nitride (PCBN) tool and 1018 steel work piece interface as functions of temperature, velocity, and normal force. Two linear models were chosen. A statistical method called membership function regression was used to determine the coefficients of these two models. The resulting model has a correlation of (Predicted Force) = 1.0445(Measured Force) with an R^2 value of 0.83. The second method was an attempt to measure friction with a measurable contact area at a range of temperatures, velocities, and normal pressures. This method rubs the end of a cylindrical rod with a concentric cylindrical pocket against a flat plate. This method caused precessions of the tool on the workpiece. As a result of this precession, plastic shear deformation friction measurements are invalid. However, Coulomb-Amonton's friction is still valid. The experiments of the PCBN-stainless steel interface found that Coulomb-Amonton's friction did not depend on temperature and velocity. In addition, no plastic shear deformation friction was identified using this method and this interface combination.

For decades models have been developed for predicting the size of the weld nugget and heat affected zones in fusion welded structures. The basis for these models is the welding heat input, which is fairly well understood for most arc welding processes. However, this traditional approach is not as straightforward for Friction Stir Welding (FSW). To date, there is no definitive relationship to quantify the heat input for FSW. An important step to establish a heat input model is to identify how FSW process parameters affect weld power. This study details the relationship between FSW process parameters and torque for three different aluminum alloys: 7075, 5083 and 2024. A quantitative weld power and heat input model is created from the torque input. The heat input model shows that decreasing the spindle speed or increasing the feed rate significantly decreases the heat input at low feed rates. At high feed rates, feed rate and spindle speed have little effect on the heat input. Process parameter versus heat input trends are verified by measurements of the weld heat affected zones. In addition, this study outlines and validates the use of a variable spindle speed test for determining torque over a broad range of parameters. The variable spindle speed test provided significant improvements over previous methods of determining torque as this new method enabled the torque to be modeled over a broad range of parameters using a minimum number of welds. The methods described in this study can be easily used to develop torque models for different alloys and materials.

Material flow in friction stir welding is largely uncharacterized due to the difficulty in material flow measurement and visualization in metals. This study investigates plasticine for use as an analog for modeling material flow in friction stir welding (FSW) of metals. Qualitative comparisons between welded plasticine and metal sections exhibit many similarities. The transient temperature response of the plasticine also shows the same qualitative behavior as welds conducted in metal. To quantify its similarity to metal, the plasticine is further analyzed through compression tests to characterize its strain, strain-rate, and temperature sensitivities. A detailed analysis is presented which defines the criteria for rigorous mechanical and thermal similarity between metals and analog materials. The mechanical response of the plasticine is quantitatively similar to many aluminum and steel alloys. In addition to the mechanical properties of the plasticine, thermal properties are measured and thermal similarity is investigated. Generally, complete thermal similarity cannot be achieved in FSW. However, given the similarities between other critical parameters, and observed qualitatively similarity, it is possible to satisfy similarity approximately, such that information can be obtained from the physical model and extrapolated to metals. Using plasticine, material flow behavior in FSW is investigated under various operating conditions. The physical model permits visualization and characterization of material flow around a suspended welding tool. Depending on operating conditions, several material flow regimes are observed, including simple extrusion with substantial tool/material slip, defect formation, a region of rotating material adjacent to the tool, and vertical deformation. Material flow and frictional heating in FSW are also investigated using a three-dimensional numerical model. Two mechanical boundary conditions are investigated, including 1) a sticking constant velocity, and 2) a slipping variable shear stress model. The constant velocity model generally over-predicts the extent of material flow in the weld region. The variable shear model predicts simple extrusion of material around the tool, and substantial tool/material slip. Additionally, the variable shear model exhibits a region of diminishing shear stress, velocity, and pressure at the back advancing side of the pin, suggesting formation of an internal void. The limited deformation, low velocities, and indication of void formation agree well with flow visualization studies using plasticine under identical operating parameters.

In order to investigate the viability of friction stir welding for use in Naval construction, mode one elastic-plastic fracture toughness of friction stir processed HSLA-65 was determined using current ASTM 1820 and BS 7448 standards. Double-sided welds were used to achieve 12.7 mm thick samples. A constant feed rate of 100 mm/min was used for all welds. To explore the effect of weld parameters on toughness, welds were produced using two rotational speeds: 340 RPM and 490 RPM. The weld centerline, advancing side hardened region (ASHR), and TMAZ/HAZ regions were sampled, in addition to un-welded parent material. All elastic-plastic fracture toughness values were thickness dependent. For welds produced at 340 RPM, toughness ranged from 33% to 75% below parent material. By increasing the rotational speed to 490 RPM, weld toughness was likewise less than the parent material, but increased 12% to 50% relative to welds produced at 340 RPM. The lowest measured toughness was in the ASHR samples for both parameters. This region of the weld exhibited mixed mode stress-strain conditions and toughness 75% and 62% less than parent material. Toughness values for all samples failed to meet qualification requirements of both ASTM 1820 and BS 7448 due to non-uniform crack extension. Irregular crack extension was caused by the through thickness change in tensile properties due to welding and the affect this had on the plastic zone size compared to the thickness. Increased weld toughness from 340 RPM to 490 RPM was attributed to microstructural differences as a result of increased rotational speed. In addition, higher crack extensions were observed in the second weld pass relative to the first for both rotational speeds. This was attributed to weld tempering of the first pass by the second. The ASHR samples exhibited the highest crack extensions. In this location, the weld microstructure consisted of Widmanstatten ferrite, a microstructure known to be detrimental to toughness.

The purpose of this study is to develop a series of response surfaces that define critical outcomes for welding in Al 7075 based on the tool geometry of a convex scrolled shoulder step spiral (CS4) friction stir processing tool. These response surfaces will be used to find critical minimums in forces which will decrease the required power input for the process. A comprehensive parameterization of the tool geometry is defined in this paper. A pilot study was performed to determine the feasibility of varying certain geometric features. Then a screening experiment eliminated those geometric features that were not as significant in determining the response surfaces. A central composite design with the five most important geometric features was used in order to develop response surfaces for nine different response variables. The nine response variables are the longitudinal, lateral and axial forces; the tool temperature, the spindle torque, the amount of flash, the presence of defects, the surface roughness and the ledge size. By using standard regression techniques, response surface equations were developed that will allow the user to optimize tool geometries based on the desired response variables. The five geometric features, the process parameters and several of their interactions were found to be highly significant in the response surfaces.

A numerical model of friction stir welding developed by T. DebRoy, R. Nandan, and others has been optimized to fit experimental data of eleven welds of 304L stainless steel at various weld feed rates and spindle speeds. Optimization was used to determine the values of five difficult-to-measure model parameters. The optimal parameter values were then correlated to the weld machine inputs. The mechanical efficiency and the coefficient of friction were not correlated with feed rate, spindle speed, or axial pressure. Tool slip was positively correlated with feed rate, negatively correlated with spindle speed, and not correlated with axial pressure. The heat partition factor was positively correlated with feed rate, negatively correlated with spindle speed, and negatively correlated with axial pressure. The heat transfer coefficient at the bottom face was positively correlated with feed rate, not correlated with spindle speed, and positively correlated with axial pressure. The above welds were instrumented with thermocouples at the mid-plane of the workpiece. Recently acquired three-dimensional temperature data indicates that the two-dimensionally optimized model does not sufficiently capture the thermal profiles in all three directions. However, optimizing the model to fit the three-dimensional data does not yield acceptable results either. Several potential sources for model improvement are identified, primarily the modeling of heat transfer at the bottom surface. It is shown that using a spatially-variable thermal contact resistance approach is more theoretically justifiable and yields better temperature predictions.

Constitutive laws commonly used to model friction stir welding have been evaluated, both qualitatively and quantitatively, and a new application of a constitutive law which can be extended to materials commonly used in FSW is presented. Existing constitutive laws have been classified as path-dependent or path-independent. Path-independent laws have been further classified according to the physical phenomena they capture: strain hardening, strain rate hardening, and/or thermal softening. Path-dependent laws can track gradients in temperature and strain rate characteristic to friction stir welding; however, path-independent laws cannot. None of the path-independent constitutive laws evaluated has been validated over the full range of strain, strain rate, and temperature in friction stir welding. Holding all parameters other than constitutive law constant in a friction stir weld model resulted in temperature differences of up to 21%. Varying locations for maximum temperature difference indicate that the constitutive laws resulted in different temperature profiles. The Sheppard and Wright law is capable of capturing saturation but incapable of capturing strain hardening with errors as large as 57% near yield. The Johnson-Cook law is capable of capturing strain hardening; however, its inability to capture saturation causes over-predictions of stress at large strains with errors as large as 37% near saturation. The Kocks and Mecking model is capable of capturing strain hardening and saturation with errors less than 5% over the entire range of plastic strain. The Sheppard and Wright and Johnson-Cook laws are incapable of capturing transients characteristic of material behavior under interrupted temperature or strain rate. The use of a state variable in the Kocks and Mecking law allows it to predict such transients. Constants for the Kocks and Mecking model for AA 5083, AA 3004, and Inconel 600 were determined from Atlas of Formability data. Constants for AA 5083 and AA 3004 were determined with the traditional Kocks and Mecking model; however, constants for Inconel 600 could not be determined without modification to the model. The temperature and strain rate combinations for Inconel 600 fell into two hardening domains: low temperatures and high strain rates exhibited twinning while high temperatures and low strain rates exhibited slip. An additional master curve was added to the Kocks and Mecking model to account for two hardening mechanisms. The errors for the Kocks and Mecking model predictions are generally within 10% for all materials analyzed.

Tooling reliability is critical to welding success in friction stir welding, but tooling fatigue is not well understood because it occurs in conditions that are often unique to friction stir welding. A fatigue study was conducted on a commonly used tooling material, H13 tool steel, using constant stress loading at temperatures between 300°C and 600°C, and the results are presented. A model is proposed accounting for temperature and stress effects on fatigue life, utilizing a two-region Arrhenius temperature model. A transition in temperature effect on fatigue life is identified. Implications of the temperature effect for friction stir welding suggest that tooling fatigue life dramatically decreases above 500°C and accelerated testing should be conducted below 500°C.

Modeling of friction stir welding (FSW) is challenging, as there are large gradients in both strain rate and temperature that must be accounted for in the constitutive law of the material being joined. Constitutive laws are most often calibrated using flow stresses from hot compression or hot torsion testing, where strain rates are much lower than those of the FSW process. As such, the current work employed a recently developed method to measure flow stresses in AA 2219-T67 at the high strain rates typical of FSW. These data were used in the development of a finite element simulation of FSW to study the effect of the new flow stress data on temperature, torque, and load predictions, compared to standard material models calibrated with hot compression or hot torsion data. It was found that load predictions were significantly better with the new material law, reducing the error with respect to experimental measurements by approximately 79%. Because heat generation during FSW is primarily a function of friction between the rapidly spinning tool and the workpiece, the choice of friction law, and associated parameters, were also studied with respect to FE model predictions. It was found that the Norton (viscoplastic) friction law was the most appropriate for modeling FSW, because its predictions were more accurate for both the transient and steady-state phases of the FSW plunge experiment. The postulated reason for the superior performance of the Norton law was its ability to account for temperature and rate sensitivity of the workpiece material sheared by the tool, while the Tresca limited Coulomb law favored contact pressure, with essentially no temperature or rate dependence of local material properties. The combination of the new flow stress data and the optimized Norton friction law resulted in a 63% overall reduction in model error, compared to the use of a standard material law and boilerplate friction parameters. The overall error was calculated as an equally weighted comparison of temperatures, torques, and forces with experimentally measured values.