While Machine learning has captured the public imagination in the consumer products such as home personal assistants and mobile devices, its application in industrial and IoT field is immense. While many ML libraries are available in open source, it is still not easy for IoT application developer to make use of them in their application. It still requires subject expert to develop ML enabled application.
Machine Learning can add value to the IoT Platform. Sensors are not always right and we don’t want to take actions on a wrong reading, due to various factors. They need to be calibrated time to time or replaced if they are faulty. But why wait for them to be faulty. The sensor can get a scorecard. Every time a faulty reading, its score value is dropped. The Management tool can trigger a re-calibration or mark the sensor for replacement, if the score of the sensor falls below a certain level. Additionally, it can choose even stop processing data from that sensor.
Here is our approach on Machine learning, and potential areas of applicability:
Vitalpointz IOT platform will offer three often used ML services in IoT and grow the list gradually. Initial ML services include: “Prediction”, “Anomaly detection” and “Classification”. With Prediction, the model can be trained to predict future values using past values of specific parameter of interest. Vitalpointz ML will provide time-series prediction, Regression based prediction as well as neural network-based prediction capabilities. Anomaly detection helps to identify any previously not-seen events, identifying outliers etc.
Vitalpointz ML will detect outliers /anomalies in both parametric and non-parametric data series hence covering most use cases. With classification, an IoT developer can estimate the probability of certain events using past data such as machinery breakdown. Vitalpotinz ML will provide several techniques such as regression-based estimation for achieving this. Vitalpointz IoT platform does not just hosts these algorithms in the cloud, it provides an easy wizard-like work flow for any developer to quickly intuitively assess various techniques simultaneously and build a suitable model with simple point and click. Once the model is built, the model can be tested in the same way. The tested model can be deployed on the cloud or optionally in the edge device.
While there is a lot of hype around Machine learning, thoughtful use-cases alongside the ML capabilities viz, “Prediction”, “Anomaly detection” and “Classification”, will enable purposeful business outcomes.