A Cloud Computing-Based Approach for Efficient Processing of Massive Machine Tool Diagnosis Data

This paper proposes a cloud computing-based approach to efficiently process the massive data produced in intelligent machine tool diagnosis flow. By collecting and extracting the vibration, power and other useful system signals during the machining operation of machine tools, the cutting process samples and cutting gap samples of machine tools can be accurately segmented, in order to construct a set of signal samples that can effectively and completely characterize the level of tool wear. We propose a visual detection method that relies on local threshold segmentation to predict tool wear status. The machine tool image is divided into several small blocks, and each image block is segmented to obtain the segmentation threshold, which is defined as the local threshold of each block. Then, the detection method scans the whole image based on the maximum local threshold among all blocks. Considering the complicated flow of visual detection and the high volume of machine tool diagnosis data, we further propose a big data processing approach which is implemented on a cloud computing architecture. By modeling the workflow of the proposed visual detection method as a directed acyclic graph, we develop a scheduling model that aims at minimizing the execution time of massive tool diagnosis data processing with available cloud computing resources. A effective metaheuristic based on search strategy of artificial bee colony is developed to solve the formulation scheduling problem. Experimental results on a cloud-based system demonstrate that, the visual detection method enhances the accuracy of tool wear detection, and the cloud-based approach significantly reduces the execution time of tool diagnosis flow by means of distributed computing.