Multiple-fault classification for hot-mix asphalt production by machine learning

To monitor the condition of the hot-mix asphalt production process, the quality and consistency of input aggregates are widely used for monitoring process variation. Current practice involves taking samples from actual production and subsequently analyzing them in the laboratory. The entire process can take up to 2 h, which, along with being expensive, is not amenable to find a single-fault pattern or multiple-fault patterns. In this paper, an intelligent hybrid classifier is proposed that can be used to recognize the hot-mix asphalt multiple-fault patterns with satisfactory accuracy. Statistical and shape features are extracted from the observation data, and the principal component analysis (PCA) is further applied to the statistical and shape features to extract effective features for the classifier. Multi-class support vector machines (MSVMs) with an adaptive mutation particle swarm optimization (AMPSO) are applied to recognize the multiple-fault patterns automatically. Simulation results show that this approach can effectively recognize multiple-fault patterns for a hot-mix asphalt production process. The proposed model has potentially good application in hot-mix asphalt fault diagnosis. The specific findings can be described in three aspects. First, by comparing to the method of using preselected parameters and the cross-validation method, the authors identified that the proposed AMPSO algorithm provides a better combination of parameters for the MSVM classifier so that the recognition rate of fault patterns is much improved. Second, the studies show that the extracted features by the PCA applied to the statistical and shape features can significantly improve the recognition accuracy. Third, this study also shows that the proposed method can deliver satisfying prediction results even with relatively small-sized training samples.