Optimization of Transient Thermography Inspection of Carbon Fiber Reinforced Plastics Panels Part 1

ABSTARCT

Transient thermography non-destructive evaluation methods are being used in aerospace industry to inspect flaws and damages for various composite materials. The purpose of this paper is to establish a set of guidelines for testing Carbon Fiber Reinforced Panels (CFRP) panels using infrared thermography. These guidelines insure that the inspection process is efficient and effective. Samples with simulated defects were made and modeled using a finite element program. Heat will be applied to the models and the temperature profiles analyzed. Along with changing the heat and time, different post-processing techniques were used to improve the method in determining defects in the sample. Once this has been optimized, actual CFRP panels with the same simulated defects were experimentally tested using the conditions from the analytical model. The analytical and experimental data was compared to insure that the testing process has been optimized. A standardized process was developed for evaluating the CFRP panels using infrared thermography.

Introduction

CFRP have been around for many years and their material properties have been improving significantly. These plastics are made up of one or more layers depending on the application requirements. There are two main types of carbon fiber patterns that use rovings, unidirectional and woven. The unidirectional CFRP, lines a layer of fibers parallel to each other, then the next layer will be the same but will have a different orientation compared to the first layer and the layers will continue to change orientation as more layers are added. In general, a simple woven pattern would have fibers going in one direction and then another set of fibers perpendicular, woven in between the first set of fibers in the same layer. Depending on the needed parameters, many layers are put on top of each other making a panel or some other desired shape. An epoxy or thermoplastic based resin will be used to bond the fibers together, and is referred to as the matrix.


A tighter weave will add stiffness and strength to the panel but will be more brittle where a loose weave will be inversely proportional. Also the type of matrix used will dictate how the panel acts in different loading conditions [1]. The manufacturing process is very complicated and there is a chance that defects shall develop internally within the CFRP. Visually the panel may look sufficient but it is critical to know that the internal structure was formed correctly. Infrared thermography can be used to help insure this quality. When a material is heated up from an external source, the heat will penetrate the sample at a constant rate. If there is a defect in the sample the rate that the heat is dispersed through the sample will be deferent at this spot affecting the surface temperature compared to the rest of the material. When the applied heat reaches the defect it will either build up at the defect or diffuse through the material faster creating a cold spot. When using an infrared camera during this process, the defect shall be shown visually on the camera’s display. However, sometimes the defects are deep compared to their size and can be hard to detect, because of this, post-processing computer programs are used to help find these defects. Optimizing this inspection process can help insure the quality of the product.

The main goal of this paper is to establish a set of guidelines for testing CFRPs using infrared thermography. This will insure that the inspection process is efficient and effective. The main objective in reaching this goal will be optimizing the parameters at which the heat applied to the sample and the most efficient way to look for defects. These are the two main parameters that can be controlled by the user with the current set-up. The supplied samples of CFRPs were experimentally tested to find values of heat and time that will identify defect areas. This gave a baseline of input data for FEA. Simulations were run to see how different material properties affected the temperature vs. time profile of the models. These models were compared with the experimental results to verify that the FEA models were correct. The experimental data was also analyzed using a post processing technique called the subtract function. This is used to refine the infrared images.

FEA Analysis

In engineering problems one simple equation is not typically adequate in solving a problem. Many times multiple sets of the same equation must be used on the same part to see how different areas of the part in question react to the applied conditions. Finite element analysis (FEA) is a method that clearly organizes multiple equations representing a modeled object and shows how the different areas interact with one another. FEA can be applied to many different conditions including: stress analysis, heat transfer, electro magnetism, and fluid flow [2]. The methods used to solve these equations can be done by hand but this would be very time consuming so a computer based program is typically used. Computer based programs like ANSYS not only save a large amount of time but can also display the results in a visual manner that is easy to understand. Many researchers have used these programs to determine the potential reactions materials will have under different conditions. Badghaish et al. used analytical models to evaluate flatbottom and embedded defects of glassed reinforced plastics. There models used a constant heat flux from the top surface and assumed that any heat transfer from other surfaces were negligible. The thermal resistance was the main parameter that was studied, which is how well one can detect the resistance of the flow of heat. This is based upon the thickness of the material which is inversely proportional to the materials thermal conductivity [3].

To completely model a composite structure would be extremely complex. Not only would each fiber have to be modeled individually but also the epoxy matrix around the fibers. Then there would be the different contact resistances for the different materials. If modeling a defect then the delamination (or other type of defect) would have to be simulated. This would require the delamination to have its own set of boundary conditions as well as the general conditions the panel has. Fortunately, this does not have to be done and a homogeneous solid can be made to represent the complex structure as long as the proper thermal properties are applied to the material [4]. If the material properties are known for the individual components but not for the entire panel, they can be combined together using

tmpFC182_thumb

where 9 is the volume fraction of the fibers, p is the density, Cp is the heat capacity and the subscripts f and m note the fibers and matrix respectively

To verify the results of the experimental testing, analytical models were made and inserted into an environment to mimic that of the experimental tests. The models were created using AutoDesk Inventor 2010 and were based off of the specifications that the actual panels were made from. Since the panels can be considered uniform though the thickness of the material the individual layers were not modeled individually. The inserts were of known thickness and size and were modeled accordingly, then inserted at depths that would represent them being between layers. Meshing was developed by using ANSYS’s automatic mesh generator which uses an automatic patch conforming sweep.

(a) The white region represents the eighth model compared to the full model with the defect represented by the blue middle square. (b) Eighth model used in ANSYS

Figure 1: (a) The white region represents the eighth model compared to the full model with the defect represented by the blue middle square. (b) Eighth model used in ANSYS

These models started out as full size models but time and space limitations prevented these models to be used. The models were cut down to smaller one defect pieces and from there to quarter and eighth size models. The eighth model is shown in Figure 1(a). The Autodesk Inventor model is based off an actual panel that is 0.49mm thick with a 6mm x 6mm x 0.2mm defect inserted 0.21mm from the bottom. The CFRP was made in 3 pieces, the part with no defect in it was one piece and then the part above and below the defect were their own separate pieces. Then the defect was modeled by itself and later all four pieces were put together into an assembly. From here the model was cut down to a 6.5mm x 6.5mm right triangle. Figure 2 shows that the scaled models reproduce the same result as the quarter scale models do within 0.01%. A panel with an inserted disk defect will transfer heat through the thickness of the material and radically outward from the center of the disk. Therefore the model can be scaled down to an eighth of its original size due to symmetry. The sides of the eighth model were also modeled to be perfectly insulated to insure accuracy.

Difference between using an eighth model compared to a quarter model yields no difference. Therefore an eighth model may be used to save computing time.

Figure 2: Difference between using an eighth model compared to a quarter model yields no difference. Therefore an eighth model may be used to save computing time.

A convergence study on the size of the mesh was studied. ANSYS has a built in mesh generator that was used where the relevance center and element size among other things can be adjusted. The default coarse mesh was used with a default element size, resulting in 12086nodes and 2455 elements; this was compared to a fine mesh with 212025nodes and 47934elements shown in Figure 3 and Figure 4 respectively. Figure 5 shows that there is little difference between the two mesh sizes, the percent difference was calculated and although the time step was initially slightly different the largest error was found to be 0.23% occurring at 0.4097sec. This is within an acceptable range allowing the coarse mesh to be used saving evaluation time.

Top view of an eighth model's mesh set to the default coarse relevance.

Figure 3: Top view of an eighth model’s mesh set to the default coarse relevance.

Top view of an eighth model's mesh set to the fine relevance.

Figure 4: Top view of an eighth model’s mesh set to the fine relevance.

Difference between using the coarse and fine mesh. There is no perceivable difference.

Figure 5: Difference between using the coarse and fine mesh. There is no perceivable difference.

Boundary conditions were reviewed to see what effect they had on the temperature of the model. There are three types of heat transfer that could be used, conduction, convection, and radiation. The model was created to mimic the actual test specimens, to do this convection and radiation effects were studied. Figure 6 is a plot of actual data compared to simulations with different types of heat transfer. When only heat flux is applied the temperature rise is very similar to the actual data but after the heating period the heat uniformly distributes throughout the sample and remains constant since there is no source for it to escape. The temperature curves of using only radiation as a source of heat loss and another curve with only convection as a source of heat loss are also plotted. Individually the amount of heat loss increases but the cooling curve is not sufficient to the actual data. However, when both these conditions are applied to the model the temperature vs. time curve of the model and of the actual experimental data is very similar. The radiation emissivity value was set to 0.97; this is the value that the experimental data is set to record as well. The temperature for this was correlated to the ambient temperature which was set to range from 22°C up to 51°C then back down to 30°C, these are estimated temperatures based on thermal couple readings of the air temperature under the hood after 5sec of heating. The convection coefficient was also based off of these air temperatures coupled with an excel sheet that calculated the natural convection of the air based on an approximate panel temperature and ambient air temperatures. The rear face of the panel was elevated from the top of the resting surface so convection was used on the rear face with ambient temperature at 22°C. The air temperature under the panel was considered to remain the same since the panel would block any direct heating, also since warm air rises the part under the panel is the coolest part.

Comparison of different boundary conditions compared to experimental data

Figure 6: Comparison of different boundary conditions compared to experimental data

The amount of heat flux used was determined by curve fitting the Ansys result temperatures to the experimental temperatures found on the panel the computer model was built to emulate. The heat flux was not considered completely constant as it is known that the halogen bulbs take a short time to heat up, so the heat flux was ramped to mimic this. The material properties of the CFRP were based on values close to the values referenced, the thermal properties of CFRP change depending on the manufacturing process and the materials used so no two panels will necessarily have the same properties so values taken were in reason.

Next post:

Previous post: