Fault Trees (FTs) for the Probabilistic Safety Analysis (PSA) of real systems suffer from the combinatorial explosion of failure sets. Then, minimal cut sets (mcs) identification is not a trivial technical issue. In this work, we transform the search of the event sets leading to system failure and the identification of the mcs into an optimization problem. We do so by hierarchically looking for the minimum combination of cut sets that can guarantee the best coverage of all the minterms that make the system fail. A multiple-population, parallel search policy based on a Differential Evolution (DE) algorithm is developed and shown to be efficient for mcs identification, on a case study considering the Airlock System (AS) of CANDU reactor.