Designing a low cost conditioned monitoring system for predictive maintenance for an Industrial Plant

Abstract

Existing maintenance strategies, corrective and preventive, are inadequate in reducing downtime. Downtime remains one of the biggest problems faced by the industrial sector, which contributes a good percentage to Uganda’s GDP. Predictive maintenance relies on monitoring machines with the help of sensors

For our prototype, the parameters considered were temperature, sound level and vibration because these will typically increase in a faulty motor-driven machines which dominate in industrial plants. The data were collected over a period of time and then used to train and test a simple machine learning algorithm based on logistic regression. Our model used sound level and temperature to determine the machine status as either normal or abnormal. The model was given historical data which it used to predict machine status and this result compared again actual machine status at the time. An accuracy of about 98% was achieved. Finally, the report recommends some actions an individual seeking to build on this project could undertake before drawing a conclusion.