To address the issue of difficulty in applying the traditional process mining on software process data due to the deficiency of activity and case attribute, this paper focus on the software process data and proposes a bilayer software process mining approach. In the mining activity layer, a weighted structured linked vector mode is proposed to vectorize the process log. The result of fuzzy clustering, which can be regarded as activity information, is determined by the average activity entropy. In the process layer, based on the heuristic relation metrics, this paper studies the non-complete cycle situation and presents the single firing sequence of loop dividing condition of log completeness, and then proposes a method to measure the affiliation of loop. The real-world data sets are used to show the effectiveness and correctness of the proposed bilayer software process mining.