The rapid emergence of digitized factories and advanced automotive manufacturing has paved the way for the integration of Internet of Things (IoT) sensors and Artificial intelligence (AI) techniques to maximize equipment uptime, that enhances manufacturing operations and support reliability. Predictive maintenance (PdM) is a key approach that leverages equipment sensor data to monitor performance abnormalities. Further, the estimation of Remaining Useful Life (RUL) aids in predicting potential or impeding failover before its actual breakdown. However, only few studies extend the work of RUL prognostics prediction to scheduling equipment maintenance, but suffer from few challenges such as overestimation of Q-values, instability during training, and inefficient performance. In addition to this, existing investigations often consider single-point RUL predictions, however in real-world scenario, many factors such as noise, distinct operating conditions of the equipment and nevertheless complex nature of equipment agening, introduces the concept of uncertainity. Hence, it becomes paramount to quantify the uncertainity of RUL, for which the authors propose non-parametric probabilistic prediction method, known as Quantile Regression (QR) integrated with Gated Recurrent Unit (GRU) and Convolutional Neural Network (CNN) for predicting RUL values of the equipment. Further, the RUL values are utilized to design better adaptive maintenance scheduling plans using the Proximal Policy Optimization (PPO) approach. The work has been tested on a prominent NASA turbofan engine dataset, which illustrates the promising performance of the proposed framework. The experimental results validate that the proposed work improves RUL prediction up to 19.08 % and minimizes maintenance schedule time up to 24.18 % in comparison to state-of-the-art approaches.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号