| Περίληψη: | In the present thesis, a numerical methodology was developed to predict the mechanical properties of unidirectional carbon fiber reinforce polymers (CFRPs) using data from NDT tests. To achieve the goal of the thesis, X-Ray CT scan data analysis was conducted, optical microscopy was utilized, multi-scale FE models were created, mechanical tests were conducted and Artificial Neural Networks were trained by utilizing the developed FE models.
Originally, UD CFRP plates with different pore content were manufactured by implementing a variety of autoclave pressures. The pressure values were the optimum, most often implemented and lowest respectively. To this end, four different types of plates of the same material were manufactured. The pores developed in the samples cut from these plates was characterized by means of X-Ray computed tomography. Prior to the final porosity quantification, a parametrical study was conducted for defining the most efficient parameters set on the software’s defect detection module. In parallel, optical microscopy was utilized for defining the large void’s dimensions in the sample with the highest pore content. By comparing the measurements obtained by the optical microscope and the XCT, the defect detection parameters were properly calibrated and validated. Small pores and micro-pores were also attributed to the analysis. At the end of this framework, porosity developed in all four samples was characterized.
Subsequently, a numerical methodology was developed which employs the pore’s characteristics. Due to the pore’s dimension high ratio between the smallest and bigger observed, the methodology was designed in three distinct levels, namely the epoxy resin with small pores, porous epoxy resin with large voids and CFRP specimens. The output of each level was the input of the consecutive. The methodology was implemented in four porosity levels’ pores characteristics resulting the mechanical properties of the CFRP laminate, namely the transverse strength and stiffness, short beam strength, flexural strength and modulus. The results in terms of mechanical behavior were validated against mechanical tests. All the matrix dominated properties appeared to be highly affected by the pores and voids leading to a degradation up to 14% of the interlaminar shear strength and roughly 20% of the in-plane shear properties.
After having validated the numerical methodology, an Artificial Neural Network was developed to link the pore’s characteristics defined by means of XCT and the mechanical properties of the porous UD CFRP laminate. To this end, a number of 30 different autoclave pressure scenarios were created excluding the 2 out of 4 actual cases, thus 30 pores dataset. After simulating the corresponding mechanical tests with the use of the numerical methodology developed previously, the ANN was trained utilizing the 30 input and output data. The two excluded cases which correspond to the most often implemented autoclave pressures, validated the accuracy of the developed ANN.
The main conclusion of the thesis is that the developed numerical methodology can accurately predict the mechanical properties of porous CFRP laminates. Additionally, the ANN based on this methodology is equally accurate and capable of predicting the degraded mechanical properties of the CFRP material using a pore’s characteristics dataset in a direct, effective and fast manner. Consequently, both two methods –after minor modifications- may be used as tools for the scopes of designing and optimizing materials saving cost and increasing the design’s effectiveness.
|
|---|
