Predictive Maintenance using Classification

I solved this problem at Paderborn University as part of an assignment of Data Science in Industrial Applications module.

Problem

Predict production of faulty parts based on historical data of ball bearing vibrations. Predictive Analysis helps business to:

• Take necessary corrective actions.
• Reduce cost as the parts rejection rate can be reduced.

Dataset

• Consists of log data from ERP system, machine data (vibration masurements) from MES system. The data distribution of normal and faulty vibrations vary greatly, so the dataset contains unbalanced data.

Tasks

• Data preprocesssing using Windowing, Fast Fourier Transform, Resampling.
• Feature generation by Spectral Transformation.
• Handle unbalanced data using oversampling the minority class (SMOTE).
• Machine learning modelling using a binary classifier.