A Graph-Attention Solution for Breakdown Prediction (GASBP)

Breakdown prediction is one of the essential steps in the maintenance of machinery devices and vehicles. The utilization of machine learning in breakdown detection is not new and has gone a long way, however, using sensory measurements of vehicles to forecast faults is still a challenge to be completely overcome. The strength of graph-based techniques in solving complex problems, and the recent success of these approaches to various modeling problems, motivated us to utilize such approaches to tackle the complex task of breakdown prediction. Thus, in this study, we propose a graph-based system that contains two main modules. The first module deals with preprocessing data which is a crucial part of the approach. In the second module, we used and adapted the GMAN technique to find the temporal aspect of the measurements and map them to the breakdown, such that we ultimately built the model for the final breakdown prediction. Our preliminary evaluation experiments on the forestry vehicles’ collected data showed promising.