Data-driven modeling of truck engine exhaust valve failures: A case study

Exhaust valve is an essential part of truck engine. Dynamic and unpredictable thermal and mechanical stress cause valves to wear prematurely, leading to increased maintenance costs. In this paper, a data-driven approach is presented to predict failures of exhaust valves of truck engines. The failure datasets of exhaust valves recorded from 13 truck engines are divided into three groups: First failure, second failure, and third or more failures. The Kaplan-Meier estimator is selected to express the distribution of survival probability of the three groups of failures. In order to find the hazard indicator, two data-mining algorithms, a wrapper and a boosting tree are applied to select parameters highly relevant to the hazard rate. A Cox proportional hazard model is used to conduct regression analysis on each selected parameter. Based on the derived hazard ratio, the time-dependent baseline hazard rate is computed. Five parametric reliability models are selected to capture the baseline hazard rate for the three groups. The value-at-risk for each group of failures is computed to express the risk at different confidence levels. Life circle of truck engine exhaust valves can be estimated.