Reliability of Smart Sensor in the Diagnosis of Unbalance and Misalignment of Electric Motors

Small and smart sensors have become popular and can be found embedded in a variety of applications. The evolution of the storage capacity, data processing, and connectivity of the sensors provided their intense use in the most different domestic and industrial applications. The exponential increase of collected information by these sensors has generated a demand for efficient data analysis techniques in order to assist the end user in decisions that are more assertive. Within this context, different manufactures launched intelligent sensors in the last year that use machine learning techniques to analyze the vibration patterns exhibited by electric motors during operation. From learning the motor normal vibration pattern, such sensors are able to identify, independently and without human interference, vibration problems by changes that happen in the operational pattern of the monitored electric motor. The purpose of this paper is to present an experimental verification of this diagnosis reliability. In this way, an agreement statistical method was used, in which one specific type of smart sensor was challenged in a “blind test” to identify two types of failure such as unbalance and misalignment. The results presented by the smart sensor were then compared before and after the fault imputation using the mentioned statistical method. To guarantee the reproducibility of the results, more than 15,000 measured data were collected in a controlled environment, inside the ambient of a research laboratory, totaling 120 operational patterns (between balanced, unbalanced, aligned, and misaligned) in a set of 60 motors of different types and sizes. The agreement index reached 92% between the generated fault and the smart sensor diagnosis.