Predictive Maintenance using Unsupervised Machine Learning

For any industrial machinery equipment, there is a need to increase operational flexibility and reduce operating costs. To achieve this objective, system engineers mainly focuses on 3 attributes of the machinery, namely reliability, maintainability and reliability. Maintenance strategy significantly improves the reliability and availability of assets and, as a result, decreases the number of unpredicted breakdowns. Recently unsupervised Machine Learning has received much attention in anomaly detection and predictive maintenance of equipments before they can fail. Unsupervised learning can help automate and improve feature engineering by extracting relevant and informative features from the data, without requiring labels or prior knowledge. In this project there are three different techniques that are applied: 1) PCA Model, 2) Auto-encoder Model and 3) LSTM Model with auto-encoder for detection of motor and compressor failures. Appropriate data analysis and feature engineering are carried out on the real-time data emanating from gas turbine. Comparative study is performed across the techniques and best performing technique is determined. It also demonstrates the model interpretability in terms of feature importance so that we can focus only on the metrics of few sensor data and monitor their health condition.

Salient Features:
– Detailed study of data analysis and feature engineering
– Data normalization and scaling of time series data
– Applying 3 different unsupervised learning methods
– Recommendation of best method for the use case
– Model interpretability in terms of feature importance

Tech Stack:
Deep Learning Frameworks, Python

Industry:
Manufacturing