Problem Use Case
A tier-1 automobile company in US wanted to predict the failure of a smoke meter device (A part of the engine
- How to predict the device failure before the error occurs like if a machine runs for 6 hours to complete the operation, sometimes the error occurs at the end of the 5th hour which fails the complete operation.
- How to predict the state of the device
- How to predict the trend of the device parameter?
- How will a model update itself and how the algorithm will work at runtime?
- They wanted an automated system where they could get all the error remedies and possible causes if an error comes.
- Canopus applied rigorous data science techniques to transform unstructured data into the structured form.
- Created a runtime model for the device failure based on the log data. Used Time Series Analysis for predicting various parameters of the device. Also created a Classification model for Error event classification.
- We designed a machine learning model to predict the parameters of the device over time. Then using those predicted parameters, we designed a machine learning model to predict the state of the device.
- By using predicted parameters and states we forecasted the device failure.
- We made a system in which the model updates itself automatically in every five minutes with the runtime data.
- Canopus designed an automated application in which the client can see the failure prediction over time. The application will visualize the predicted parameter values over time with graphs and data tables and also the possible causes and remedies regarding the errors.