Healthmon: um novo sistema para prover manutenção preventiva de máquinas através da identificação de evoluções de falhas em conjunto de séries temporais

Abstract

In the context of Industry 4.0, monitoring machine degradation to anticipate possible failures represents a significant challenge. That task is especially important when the costs imposed by maintenance and stopping production processes are high. Nowadays, many preventive maintenance techniques employ supervised or unsupervised machine learning algorithms. However, the definition of which variables should be processed by those algorithms is not a simple task, being crucial for the success of the proposed technique. Given this perspective, we consider whether unsupervised algorithms combined with the decomposition of time series can contribute to improve the health of monitoring machines. In this research, we propose HealthMon, which is a new approach, whose function is to calculate a machine integrity index based on sensor measurements. HealthMon extracts time series from those sensors, which are decomposed in a new and unsupervised way, in addition to being used to calculate the integrity index of that machine. That integrity index is related to the degradation of the machine considered, being useful, therefore, to optimize the maintenance schedule of the machine. This research advances the state of art in the following points: (i) proposes a global index of health, which offers a more direct and intuitive view of machine degradation; (ii) elaborates an approach capable of estimating the integrity index of a machine, while only basic knowledge about the (internal) operation of the machine is necessary; (iii) offers an extensive range of applicability, since it is scalable so that it can be used in any type of vibrating or rotating machine. Our method was evaluated through extensive simulations on an induction motor. The results have shown that the degradation can be effectively detected under various input workloads. Besides, it is worth noting that HealthMon was evaluated through the usage of real data. In this regard, it’s interesting to highlight that, in both evaluations, we found that the results obtained were promising with regard to our hypothesis.