The Internet Of Things is enabling a transformation that makes it possible to gather and analyze data across machines, enabling faster, more flexible, and more efficient processes to execute at higher-quality at reduced costs. But how do you start? What are the steps you have to take to go from data collection to improving your processes? Let’s describe a real case we implemented to answer these questions.

We believe complexity should be hidden from the users, we believe you can apply machine learning without being a data scientists, we believe you can collect data without being a data engineer and finally we believe you can create your own dashboards, actions and control your processes without being a programmer. We do this by putting you in control of the tools.

Let’s present you with a real use case we implemented. It is a simple use case, a customer asked us if we could help monitor thousands of remote devices that are pressurized. These devices have the capability of sending their current status every hour. The question is which ones are leaking and need attention by a field service engineer.

 

We started with collecting the information, like many sensors and machines nowadays the customer can already configure the pressurized device to forward the data securely to the MEUNGO application. Actually, this customer started their Internet of Things project enhancing their own hardware to communicate the status, while doing this they used the MEUNGO software to test and verify the communication. Out of the box the MEUNGO software allows users to add and change data sources (sensors), how they communicate and how to integrate these data sources into one device type. We believe we live in a changing world and therefor allow the user to change the definition of the world, we do not program it, you as a user configure the system.

After connecting the devices and verifying the data using user defined dashboards the customer was ready to create some information out of the data. The challenge here is the amount of devices, not always knowing the desired pressure and the fact that some of these devices are exposed to outside temperature that influence the pressure. Simply detecting a lower pressure is not sufficient because that could be because of lower outside temperature, manually checking for a trend in one the user defined plots does not scale well and is error-prone.

The solution the customer selected was to invoke the simplest machine learning algorithm one can execute on a time series. The customer decided to configure the system to train a linear trend algorithm every day using a week of history. The trend of the estimated line indicates if the pressure over the course of a week is increasing (larger than zero), is stable (zero) or is leaking (less than zero). Next, the user defined a rule to be triggered only if the trend is less than zero, the actions associated with the rule are to record the event and create a work order in their field service software.

The customer was able to go from collecting data to action in less than a day and continuous adding actions based on progressive insight. On top of that the solution allowed the customer to improve their Service Level Agreement by guaranteeing the presence of pressurized devices at a lower cost.