Improving Production Sustainably with Data Science: The Monitoring API for Data Scientists

Pubilshed on December 10, 2020 From Vanessa Kluge

Everyone is talking about predictive maintenance. Popular search engines currently score over 5 million hits. Predictive maintenance is one of the most common use cases at conferences and events in the mechanical engineering environment. The range of best practices from many providers is impressive, but the path to get there can be challenging.

Why? Because several factors determine the successful implementation of predictive maintenance applications. These include, among other things:

1.    Data quality and scope
2.    Data visualization
3.    Intelligent data integration
4.    Data analysis competence
5.    Practical process knowledge

Companies that want to offer predictive maintenance as an extended service for their products need a clear strategic commitment as well as resources (competence, budget, capacities) and the right tools. 
This is exactly where Kontron AIS comes in with its digital IIoT service solution EquipmentCloud®. It enables machine manufacturers to record and evaluate their machine data (faults, process data, status, units produced) in real time and build up a data history to create the foundation for data-based business models. To apply machine learning algorithms, data scientists rely on the Python programming language and specialized tools such as Scikit-Learn, Tensorflow, Pandas, Azure ML Services and Jupyter Notebook. These development tool enable in-depth data analyses including statistical evaluations, correlations, superimposition of diagrams and visualizations.

Since process data is of significant, if not decisive importance, for predictive maintenance applications, the EquipmentCloud® serves both as a relevant source of information and storage, as well as for productive application of the insights gained.
In order to facilitate the interoperability of the various applications and thus also data retrieval for data scientists, the existing monitoring REST API has been fundamentally expanded. 
The REST API allows authenticated users to query stored consolidated data such as alarms, production volumes, status and process values using GET requests. Authentication is carried out using multiple credentials such as username, password, and customer ID. 

For more advanced use, a separate Python module was developed as a REST API wrapper. The Python module PyEqCloud acts as a REST API wrapper in a sense that it makes the standardized functionalities of the EquipmentCloud® REST API accessible for working with Python as a programming language and provides directly usable functions.
The module was published as a project on in PyPi (Python Package Index) and can be installed and used with an MIT license as open source in a corresponding Python environment. Data scientists therefore benefit from data streaming and the further processing of data using the Python module PyEqCloud.

This results in a basic toolset and the following proven workflow that can be used, for example, to identify defective components:

1.    Machine coupling
2.    EquipmentCloud® as storage
3.    Data retrieval using REST API
4.    Analysis locally or in the EquipmentCloud®
5.    Local or cloud training
6.    Productive use in the EquipmentCloud®
For this purpose, alarm and event data are queried over a period of several months. These are viewed and processed statistically to get a feeling for the data. To analyze possible influences on the data, correlation analyses of faults and quality parameters are then carried out.
That is how possible factors influencing the performance of motors can be identified at an early stage during production. In this way, faults and alarms can already be triggered preventively and rejects can be significantly reduced. This is a method of establishing an early warning system. 
In order to adjust this early warning system as precisely as possible, a trained model is tested on the basis of the identified parameters and classifications are derived 
Finally, the aim is to test the trained models on productive data streams as live visualizations in the EquipmentCloud® in order to continuously develop the models. 
The EquipmentCloud®, REST API and the Python module are therefore the key to predictive maintenance.
Are you a data scientist yourself or do you have specific application scenarios that you would like to put into practice? 

 Get started in just a few steps!

Download code and manual on GitHub