Streaming Data Analytics

PROBLEM: Building preconfigured machine learning based solutions on Nebbiolotech platform for anomaly detection and predictive maintenance

SOLUTION: Our preconfigured UDF chains uses artificial intelligence to detect anomalies to prevent breakdown of your services by performing appropriate analytics on your streaming data at the edge itself so that you are immediately alerted when you need to take remedial action and know what action will be appropriate at that time. Our anomaly detection methods provide a ready solution generating alerts at different severity levels so that effective action can be taken by the production managers. We also predict the future performance of different machine parameters based on past data.

Anomaly Detection and Predictive Analytics

PROBLEM: We were engaged by Nebbiolo to assist with a problem from a large Automobile manufacturer with Robots on the shop floor to analyze vast amounts of sensor data to trap anomalies. A history of anomalous events was not available to facilitate detection.

SOLUTION: Analytics Plus captured and analyzed sensor data from Kuka Robots using sensors deployed by Nebbiolo. We used unsupervised data cleaning algorithms based on clustering (K-means/Dbscan) to detect anomalies and eliminate sensor error or missing data. Prediction algorithms like HoltWinters, EWMA, ARIMA were used to predict future movement of data under normal conditions and deviations from the norm was used to predict anomalies. An App was built on the Nebbiolo platform allowing users to use/ set up pipelines to analyze data from sensors.

BENEFIT: The solution developed on test lab robots was able to capture genuine anomalies in real time and predict future robot arm movements based on past data. This predicts anomalies in advance and thus allows predictive maintenance and root cause analysis to be done on shop floors.

Tools Used: Influxdb, Kapacitor, Grafana, Python.