Agnomalies | Devpost

Inspiration

We wanted to find a solution that would help farmers with predictive maintenance. Looking at the data provided, we had trouble finding when a failure actually occurred, let alone how to predict it. Doing some research online, we found some research about how anomaly detection has been shown to be somewhat effective in predicting machine failure. We wanted to build a tool that could take streamed data from machines, detect anomalies, and use this information to inform farmers that maintenance may be required.

What it does

Takes streamed data, determine if an anomaly is present, store raw data and anomaly scores, and display on a website.

How we built it

We used machine data to generate signals similar to how it would be sent by a machine in operation. This data stream was sent to Amazon Kinesis, where we used Random Cut Forests to generate anomaly scores for each component and the two major subsystems (engine and equipment). The data is then processed by an Lambda function which processes it and places in to a DynamoDB database. We also have a website that makes calls to a Lambda function to get the most up to date machine data/anomaly score. If an anomaly occurs, another lambda function is called and an SNS is sent to the user.