A new method to achieve dynamic heat input monitoring in robotic belt grinding of Inconel 718

Abstract

Robotic belt grinding is one of the most effective methods to process difficult-to-machine materials, such as Inconel 718. However, unpredictable heat generated in grinding area may destroy the surface quality of components. Computationally intensive numerical methods and inaccurate analytical methods cannot reliably obtain the heat input in grinding processes online. To this end, we propose a novel method based on multi-sensor and machine learning techniques to achieve online dynamic heat input monitoring. Firstly, comparison of the dynamic and static heat input is performed for illustrating the necessity of online heat input monitoring. Secondly, through associating the grinding signals (sound and force) with the heat input and developed feature selection method, a BADS-LSSVM (Bayesian adaptive direct research-least squares support vector machine) based model is proposed to predict the heat input. Test results show that the proposed method has a mean accuracy of no less than 96.7 %, a computed temperature error of ±6 °C in a complete grinding pass; and takes about 0.6 s for each calculation. With this new method, the real-time heat input monitoring is achieved and the subsequent thermal control of robotic belt grinding can be further conducted in future.