Predictive Maintenance using AI for a leading German Semiconductor manufacturing company
The client is a leading semiconductor manufacturing company based in Germany, with a very strong presence in the semiconductor market. It offers a gamut of solutions ranging from semiconductor manufacturing optics to photomask solutions.
The client was keen on reducing cost incurred due to machine failures and downtime. It wanted to drive a root cause analysis to understand how operational performance metrics like cycle time are getting impacted by external factors such as pressure, temperature etc.
Historical data provided by client for analysis was fraught with many imperfections. Preliminary analysis revealed the following anomalies that increased the complexities in the process:
- The client had a manual recording process which left the end database highly inaccurate.
- There were high number of missing values and outliers.
- Imbalanced dataset and feature duplication intensified the complexities.
- Visualization capabilities for tracking and monitoring were required to be built from scratch.
MS Azure Table, Azure Machine Learning Studio, MS Power BI Dashboard
- Machine learning-based imputation techniques were used contextually to address missing value and outlier issues, which helped create a balanced dataset.
- Feature engineering was applied to reduce the dimensions
- Decision tree model was conceptualized to provide condition status of the machine either as Pass or Fail based on parameters such as Pressure and Temperature of sensors.
- The output from machine learning model was fed to the sensors with specific pressure and temperature conditions. Based on the fed values, the sensors would trigger alarm and alert system engineers for necessary corrective action.
- Stable maintenance cycle increased productivity and significantly reduced machine failures.
- By providing the root cause of fault occurrence in machines, the predictive maintenance solution reinforced the efficiency of operational planning. Increase in machine durability resulted in reduction in idle time.
- Remarkably, the solution enhanced the accuracy of preventive maintenance to 86% as compared to the previous approach.
- This solution helped client to save millions of dollars.