Transport infrastructure is a key asset in the global development of a society. Its improvement, both from the technologies and materials point of view, is a precondition to achieve sustainable and resilient progress. Under these conditions, it is necessary to investigate the feasibility of using waste materials (also called marginal materials) into a pavement mix composition in order to accomplish two main goals: adding useful life to a material that otherwise would be just a waste and saving on virgin materials supply. The presented paper aims at investigating the feasibility of using industrial waste silica fume (SF) as a surrogate filler instead of Ordinary Portland Cement (OPC). Mixtures were produced with different percentages of SF and OPC and their performance was investigated by using the results of Marshall and Indirect Tensile Strength (ITS) tests. Laboratory testing requires not only highly qualified technicians but also multiple samples in order to define functional relationships between the variables involved. In this regard, soft-computing techniques can be useful in reducing this workload and the resulting costs by identifying the aforementioned relationships by means of artificial intelligence. Therefore, the experimental data collected have been processed using Shallow Neural Networks (SNNs) that provided predictive models of the mixtures' mechanical and volumetric parameters. Resampling and synthetic data generation techniques successfully addressed the difficulties caused by the relatively small dataset size. Results showed that the use of SF resulted in mixture performance comparable to that achieved by mixtures produced using OPC, occasionally even better. In addition, the proposed neural model performed remarkably well and thus could be used in the asphalt mixture optimization without the need for additional laboratory tests.
