Overview / Usage

The impact of unscheduled equipment downtime can be detrimental for any business. It's critical to keep field equipment running to maximize utilization and performance, and to minimize costly, unscheduled downtime. Early identification of issues can help allocate limited maintenance resources in a cost-effective way and enhance quality and supply chain processes. This predictive maintenance project provide a data science project from data ingestion, feature engineering, model building, and model operationalization and deployment. Most businesses are interested in predicting when these problems might arise to proactively prevent them before they occur. The goal is to reduce the costs by reducing downtime and possibly increase safety. The business problem is to predict issues that are caused by component failures. The business question is "What's the probability that a machine goes down due to failure of a component?"

Methodology / Approach

A machine learning algorithm is used to create the predictive model. The model is trained on historical data that's collected from machines.

1. Prepare and collect your raw data
   
2. Create model features and target label :
   

• Machines: Features that differentiate each machine, such as age and model.
  
• Error: The error log contains non-breaking errors that are thrown while the machine is still operational.
  
• Maintenance: The maintenance log contains both scheduled and unscheduled maintenance records.
  
• Telemetry: The telemetry data consists of time series measurements from multiple sensors within each machine.
  
• Failures: Failures correspond to component replacements within the maintenance log.
  

3) Model Building & Evaluation
4) Deploy the stored model as service on an hosted web service
5) Building web dashboard ( for data reports ) & notification system ( for alert & warning )

Technologies Used

Azure ML Studio, PowerBI, TensorFlow, Movidius

Collaborators