Multimodal operating status prediction for interpretable health assessment of autonomous mining trucks

Health assessment is a critical aspect of industrial prognostic and health management (PHM). Evaluating the health status of autonomous mining trucks (AMTs) is crucial for enhancing the operational efficiency and safety of mining operations. However, the complex structural composition of AMTs, dynamic operational environments, and the lack of health benchmarks pose significant challenges to this task. To this end, this study proposes an interpretable health assessment framework based on multimodal operating status prediction for AMTs. First, a Transformer architecture is employed to establish a multi-quantile prediction model, which integrates multimodal data to predict the physical variables characterizing the key system operating status of AMTs. Since these predictions serve as references for health assessment, the assessment process is made interpretable. The model also considers the impact of abnormal positioning in mining environments. Subsequently, health assessment models for three key systems of AMT are developed using quantile prediction results, an optimized weighted loss function, and expert experience-labeled data. Last, real-world experimental results demonstrate that compared with three state-of-the-art methods, our approach achieves the best performance, with an average root mean square error (RMSE) reduction of 39.9% and an average score reduction of 35.7%, indicating the effectiveness of the proposed health assessment framework.