The detection of electrical failures and more specifically electric arcs faults detection becomes absolutely necessary because of their dangerousness. Unlike the main methods in the literature based on frequency analysis, we propose in this paper a method of detection based on CNN models (LeNet5-28*28 and 64*64 images). Series arc faults are produced in a circuit composed of a 270 Volt DC supply voltage and mainly resistive loads. The line current is recorded (dataset composed of about 11000 signatures with and without arc faults). We also consider the case of load variation without arc faults. The selected sections of the current signals are then transformed into 2D matrixes (images). Then, the network is trained, validated and tested using the dataset. The performance of this method is also studied and discussed. More precisely, the detection results are presented using a confusion matrix. Experimental results show that the method we propose can effectively detect arcing faults.