Advanced search
Start date

Machine IA: online predictive maintenance system on electrical machines using machine learning techniques and an autonomous sensor network based on energy harvesting technology


Mechanical equipment is inevitably subject to wear, which can lead to defects and breakages. In industries, breaking a machine can cost a lot of production in terms of time and quality of service. Thus, in order for productivity in manufacturing processes, which typically have a significant diversity of equipment and machinery, to achieve positive results, the entire park must be maintained in the best possible operating condition. In this way, machines and equipment routinely undergo repairs, inspections, replacement of parts and oils, cleaning and corrections, often on a scheduled basis within what is commonly called maintenance.Maintenance of machinery and equipment is often subdivided into two categories, namely: corrective maintenance, which involves repairs after equipment failure; and preventive maintenance, which involves a set of measures to predict and prevent failures. The focus of this feasibility research is preventive maintenance, more specifically predictive preventive maintenance. Unlike systematic preventive maintenance, which is simply based on uptime and statistical criteria, predictive preventive maintenance considers the analysis of data that will be constantly monitored on machines. The use of Machine Learning techniques in predictive maintenance (online) has taken up space in academia and industry as an important tool for cost optimization and increased efficiency and productivity. In this context, vibration and oil analysis are the main parameters used to detect faults. Vibration analysis is used as a non-intrusive technique where mechanical system response can be measured - for this reason, vibration variation will be considered in the predictive maintenance system to be developed in this research. Specifically, Machine Learning addresses the study and development of methods and computational algorithms capable of learning and improving from the data itself, aiming to make predictions - insights. As a branch of Artificial Intelligence, Machine Learning techniques act more intelligently than those formed from strictly static instructions, assisting in decision making directly from the analysis of problem data. Machine Learning algorithms can recognize patterns in data from two elementary types of learning: supervised learning and unsupervised learning. These two learning paradigms provide, respectively, foundations for the study and development of so-called data classifiers and groupers. In this work, will be investigated primarily the data classifiers, seeking to predict the break or failure of mechanical and electrical elements. (AU)