Topology optimization (TO) makes it possible to obtain new structures for electrical machines. The sensitivity-based method, which can cope with some constraint conditions, is suitable for large-scale three-dimensional TO. However, if the material density is defined by unknown variables in TO, elements with intermediate density (grayscale) occasionally appear. The grayscale cannot clearly show the material allocation within its finite element. Thus, we propose a sigmoid-based filtering function to suppress the generation of grayscale. Moreover, because the constraint condition can be simply taken into consideration, sequential linear programming is occasionally utilized as a topology optimizer. However, the convergence characteristics frequently oscillate and are strongly dependent on the move limit that controls the maximum intensity of the correction vector. To overcome this numerical difficulty, we propose an identification technique for the determination of a quasi-optimal move limit (QOML). This paper demonstrates the performance of both the mathematical function filtering grayscale and QOML.