Feature attribution methods (AMs) are a simple means to provide explanations for the predictions of black-box models such as neural networks. Due to their conceptual differences, the numerous different methods, however, yield ambiguous explanations. While this allows for obtaining different insights into the model, it also complicates the decision regarding which method to adopt. This article summarizes the current state of the art regarding AMs, which includes the requirements and desiderata of the methods themselves as well as the properties of their explanations. Based on a survey of existing methods, a representative subset consisting of the δ-sensitivity index, permutation feature importance, variance-based feature importance in artificial neural networks and DeepSHAP, is described in greater detail and, for the first time, benchmarked in a regression context. Specifically for this purpose, a new verification strategy for model-specific AMs is proposed. As expected, the explanations’ agreement with the intuition and among each other clearly depends on the AMs’ properties. This has two implications. First, careful reasoning about the selection of an AM is required. Second, it is recommended to apply multiple AMs and combine their insights in order to reduce the model’s opacity even further.

Internal cooling structures for gas turbine engines are becoming more complicated to push the hot gas temperature as high as possible, which, however, allows particulates drawn into the coolant air to be more readily to deposit within these passages and thus greatly affect their flow loss and thermal performance. In this study, internal swirl cooling and jet impingement cooling subjected to particulate deposition were evaluated and compared using a conjugate heat transfer method, with an emphasis on the thermal effects of the insulative deposits. To accomplish the goal, an unsteady conjugate mesh morphing simulation framework was developed and validated, which involved particle tracking in an unsteady fluid flow, particle–wall interaction modeling, conjugate mesh morphing of both fluid and solid domains, and a deposit identification method. The swirl and the jet impingement cooling configurations modeled the internal cooling passage for the leading-edge region of a turbine blade and were investigated in a dust-laden coolant environment at real engine conditions. Coupling effects between the dynamic deposition process and the unsteady flow inside the two cooling channels were examined and the insulative effects of the deposits were quantified by comparing the temperatures on the external and internal surfaces of the metal channel walls, as well as on the deposit layers. Results demonstrated the ability of the newly developed, unsteady conjugate simulation framework to identify the deposits from the original bare wall surface and to predict the insulation effects of the deposits in the dynamic deposition process. The dust almost covered the entire impingement channel, while deposits were only seen in the vicinity of the jets in the swirl channel. Despite this, a dramatical decrease of convection heat transfer was found in the swirl channel because the swirling flow was sensitive to the interruption of the deposits. In contrast, the deposits improved the heat transfer rate in the impingement channel. When the thermal effects of the deposit layer were taken into account, the wall temperatures of both two cooling geometries were substantially elevated, exceeding the allowable temperature of the metal material. Due to the denser deposit coverage, the impingement channel wall had a greater temperature increase than the swirl channel. In terms of flow loss, the presence of the deposits inhibited the swirl intensity by interrupting the swirling flow and thus reduced the friction loss, whereas the pressure loss was improved by the deposits in the impingement cooling.