Predictive maintenance on an elevator system using machine learning

The aim of the project is to design and construct a predictive maintenance system on an elevator system using machine learning. Three objectives are set for the project for the system to be archived. The first objective is to develop the machine learning technique to monitor the health of the elevator system. The elevator system chosen for the predictive maintenance system was permanent magnet synchronous motor traction elevator. The PMSM data set was proceeded for the data analysis. Smoothening of the data were completed as there were too many peak data in the data set. Using the smoothen data set, threshold was created for the classification of the output (health condition) by comparing with the “time” parameter data. Once the output has been classified and tabulated into the data set, the completed information data set will be transferred into the model for training purpose. The second objective was to design a prototype framework of an elevator for data collection. Arduino Uno was chosen as the microcontroller for the elevator prototype. DC motor was selected to representing the elevator motor that drives the elevator car. Two sensors: LM35 and Encoder Sensor Module were selected for the data capturing objective. LM35 is capturing the temperature data ad Encoder Sensor Module will capturing the rotation per minute data form the DC motor. The data collected will be compiled into a file before transferring it to the MATLAB processing. The last objective is to evaluate the performance of efficiency of the system. Total of 5 testing were conducted for the implemented system. The first three was about the setting of training model, the result was Fine KNN algorithm has the most accuracy of 93.8%. The fourth testing was conducted on checking the prediction ability of the trained model. The analysis shows the trained model maintained its accuracy even when extending the range of time for prediction. The fifth testing is about unbiased prediction of the trained model. The final result of the unbiased prediction accuracy was 95.5%.