Machine Learning, Data Science, Big Data and IoT are actual buzzwords on most of the industrial fields. The number of applications is uncountable, and new opportunities, algorithms and tools appears every day. On the mining industry, two fields are on the scope of machine learning applications: Process Optimization and Condition based Maintenance (CBM). CBM requires a set of indicators to show that an equipment is losing performance or is about to fail. To build those indicators, sensors that can measure some specific variables are key to detect and/or predict the appearance of some faults. Those variables are commonly vibration, temperature and acoustics. On the classical approach, an expert analyst studies the data of every measurement point and fills a form reporting the condition of the asset. The machine learning approach uses the information obtained from the sensors to feed an algorithm that was pre-trained to solve the requested task, to perform its own analysis, in a fraction of a second. The algorithm learns to detect and identify some failures, based on examples obtained from real assets and laboratory tests. In this article, examples of failure detection and identification on bearings are presented. The algorithm uses raw vibration data as input and computes a set of statistical features that are used to train a classifier1. Results show an accuracy over than 97% on different public and proprietary data sets. This solution allows the creation of Smart Assets and IIoT applications to monitor and solve many of the current reliability issues as unplanned maintenance.
展开▼
