Application of Machine Learning in Generating Load Profile Transitions for a Gear Pair

A gear is an essential component of a mechanical transmission system. Its design and reliability have a great influence on the system service life. The design of a gear pair is generally done using various design criteria and constraints. At the design stage, since the future operating profile in terms of terrains, load transitions and speeds are unknown, one method for extending the life of a gear teeth is to use a high safety factor. Unfortunately, this strategy is not always acceptable because it not only adds unnecessary weight but also increases the cost. In this paper, a machine learning technique is used to generate operating profiles that a gear pair can experience in the future. The proposed method is based on the N-grams algorithm which is extensively used in Natural Language Processing to predict the next word on the basis of the order of the previous words. In the current work, three different terrains are considered, each with a range of torque from 1000 Nm to 9000 Nm. This range of torques is further segmented into five subranges, where each subrange of torques represents a state. A state in a particular terrain has its own random exponentially distributed residence time. The future path would be the combination of different load states. The generation of sequence of states in a terrain is accomplished using N-gram. The method proposed is practical and applicable at the design as well as for prediction of residual useful life of the gear pair.