Nowadays, neural networks have been effective in solving many deep learning problems. Deep neural networks are used in various applications such as natural language processing, speech recognition, and computer vision. Fault occurrence is one of the most critical issues that threatens the operation of computer systems and affects the performance of deep neural networks. Many researchers are striving to identify the types and locations of faults through fault simulation in order to reduce the vulnerability of neural networks. The presence of numerous potential faults in modern systems poses a challenge for simulation-based fault injection since designers need to perform both fault-free simulations and thousands of faulty simulations. Fault simulation and the identification of critical faults require the injection of thousands of faults, which is often challenging and time-consuming. In this thesis, a method is proposed to accelerate fault injection simulation in convolutional neural networks, allowing for faster evaluation of their vulnerability. The implementation of this method was carried out in the PYTHON environment using the PYTORCH library. Using this approach, the fault injection time was significantly reduced, and the results from the experiments demonstrate the efficiency and acceleration of vulnerability analysis in convolutional neural networks.