Reducing cost with MEMS sensor and improving performance of classifier using probabilistic voting method

Abstract When developing a solution for fault diagnosis, cost is a critical factor that must be considered to ensure a practical and feasible final product. To reduce expenses, the authors conducted a feasibility study using a low-cost MEMS accelerometer in conjunction with an Arduino Uno. This study aimed to determine the effect of sample length on fault classification by varying it from 50 to 5000 to identify the optimal value for this particular application. After finalizing the parameters, the vibration signal was acquired using the MEMS sensor and Arduino Uno. The resulting classification accuracy using the decision tree was 90.2%, a satisfactory result that can be further improved for industrial applications. To enhance the accuracy of decision tree classifiers, the authors proposed a novel approach known as the probabilistic voting method. By implementing this method and utilizing an Arduino Uno and MEMS sensor (ADXL335), they were able to drastically cut down costs while achieving remarkable classification accuracy. The implementation of the probabilistic voting method further elevated the accuracy to an astounding 98.5%, setting a new benchmark in the field.