GMM Based Fault Signature Estimation of Electromechanical Machines for Small and Medium-Sized Enterprises in IoT Environment

Abstract

Small and medium sized enterprises (SMEs) form backbone of a nation’s economy. Implementation of technologies like Internet of Things (IoT), however, is a challenge for majority of them as the conventional solution requires a lot of investment. Thus, financially restricted SMEs, especially in developing nations, remain aloof from leveraging the benefits of the technology. Resorting to affordable devices such as low-cost sensors, actuators, processors, servers, and network technologies etc., pose challenges like low memory, low computation power, less transmission power, low data transfer rate, and limited network bandwidth. Consequently, there arises a need to develop IoT based solutions that cater to these challenges so that low budget SMEs are also able to benefit from IoT’s umpteen advantages. This paper proposes an affordable IoT based framework for health status monitoring of machines in SMEs keeping the limitations imposed by low cost IoT devices as centre of the solution. The scope of the present research is limited to monitoring the health status of the electromechanical rotating machines only. Four types of commonly occurring faults in the machines at different rotating speeds are investigated using acoustic signals generated within the machines. Mahalanobis distance and Gaussian mixture model (GMM) have been employed for the analysis of the acoustic signals for estimating the unique fault dependent signatures. GMM works satisfactorily with smaller datasets and requires lesser amount of computational power in comparison to machine learning based algorithms. The investigations have showed that GMM may be effectively used in resource constrained SMEs deploying affordable IoT devices for predictive maintenance.