The system coupled automated Time Series Analysis with Data Dashboards to leverage Predictive Maintenance algorithms coupled with Human Oversight for its production line.
A large heavy equipment manufacturer had deployed IoT-based sensors on key equipment deployed on the production line. The sensor data was manually analyzed to detect and provide early warning of equipment failure.
A need was felt to automate the analysis of the large volume and type of data captured via the various sensors such that automated alerts could be generated.
If the analysis detects warning signals, Data should also be displayed on intuitive dashboards. It will allow employees to monitor the equipment in real time.
Machine learning based analysis of historical sensor data can detect the need for preventive maintenance before the same may be evident to human operators. Thereby allowing one to proactively schedule maintenance to prevent costly outages. The resultant avoidance of last-minute repairs and part replacements goes a long way towards achieving maximum operational efficiencies on the production line.
Sensor data analysis is a key process for the detection of anomalous readings. Hence, it is a key indicator of abnormal operating conditions of the equipment under review. The source of the abnormalities could be environments or inherent in the equipment itself – either way, such anomalous operating conditions are often the precursors of equipment failures. Therefore, early detection and suitable escalation to human operators for investigation are critical to smooth operations.
With the proliferation of IoT-based sensors, it may seem that we have achieved the nirvana stage. And that all data will automatically become available for analysis and display. However, a lot of pre-processing and data cleansing activities need to be carried out, at scale, in order to unlock this potential value. The presence of a wide variety of sensors and data collection mechanisms makes this doubly complicated. An experienced partner like us can help deploy the data collection, cleansing, and storage mechanisms that form the base for further machine learning in predictive maintenance and analytics.
We deployed the following technology to create an effective predictive maintenance alert through machine learning and IoT:
Developed a feature extractor for the time series data on Current, Temperature, and Vibration readings from various sensors using Python packages (NumPy, Pandas).
Deployed a rules engine based on time series analysis to automatically detect problematic events of interest and to generate alerts based on the detected events. The alerts were automatically sent as mobile app notifications and as emails.
The resultant model was deployed to the customer’s on-premises infrastructure and the generated alerts, and dashboard graphs were pushed to a React-frontend website, deployed on the cloud.
the Next-Level Potential of
Human and Machine Intelligence