Developed a framework for data integration and failure detection in Sensors/Internet of Things (IoT) installed on various components of off shore Wind Turbines.
•A neural network classifier with 2 layers is trained with healthy and damaged data for detection, and true positive detection accuracy for random testing stands at 95%. Project is fully funded by Maryland State Government and our work paved way for industrial partnership with IBM smarter planet. Currently a proposal is being prepared for NSF grant for continual research.
Log Analysis Using Apache Spark and MatLab
• Designed a service to parse through 100,000 apache web server log lines and extract the features like IPAddress, URI, time zone, username, response code etc. and save them into CSV file using Python API in Apache Spark and deployed in AWS EC2 Cluster.
• Trained a single node Neural Network with sigmoid internal function in MATLAB to classify the response code for every hit depending on username and URI user is trying to access. A true positive accuracy of 86% is achieved, and this model can be extended to any classification problem deemed fit.
• Technology Stack: Hadoop, HBase, Hive, AWS (EC2, S3), Apache Spark, Databricks, Python, Jupyter, MatLab