Data-driven decision taking in IoT projects

Our customers have very different motivations for connecting their devices. These include, for example:

• (Remote) monitoring of the devices.
• Manual (remote) control.
• Intelligent management, e.g. optimization of energy consumption
• Increasing comfort of end customers, especially in smart homes
• Shift to new business models, e.g., product-as-a-service
• Individualization

Understanding business processes

At the very beginning, our customers have a specific problem or goal for which they would like to find a solution with us. This point can be at the beginning of a product development or at a later stage when data is already available. Together we work to understand all aspects as comprehensively as possible. The better the problem can be described, the better we can find suitable options for implementation. The term “suitable” has various dimensions - the main ones are technical feasibility and economic sense.

Once we understand the problem, we start planning and defining the framework. Understanding the business processes is the foundation to overcome analytical and technical challenges.

Use cases in the Internet of Things

A typical data-centric use case in our IoT projects is predictive maintenance (Predictive Maintenance) . Here, maintenance requirements are no longer defined classically on the basis of fixed units of measurement (such as months, useful life, or mileage). Instead, the actual condition of the product is to be considered.

The aim is to optimize the timing of maintenance. The maintenance intervals are maximized, but in a precise way that prevents failures and minimizes downtime. Here it is also important to order needed spare parts in time or to incorporate spare equipment.

Data-driven decisions

To successfully implement data-driven decisions, all framework conditions must be considered. Using predictive maintenance as an example, the following questions are particularly important:

• What is the usual product lifetime?
• What currently triggers a maintenance need?
• What is the general contract situation regarding warranty and maintenance?
• Who performs the maintenance?
• How frequent are breakdowns (despite maintenance plan)?
• What are the costs that failures cause and who bears the costs?

To drive this development, there must be sufficient potential for maintenance management improvements. In practical terms, this means that undesirable situations occur, such as unnecessary or delayed maintenance, failures, or accumulations of warranty issues.

It is often the case that scenarios emerge from the implementation of a use case that can be implemented on the basis of the collected data. The customer may want to establish new business models based on the holistic maintenance process. In Product-as-a-Service (PaaS) or the Equipment-as-a-Service used in Industrial IoT (IIoT), the customer pays for the use. He does not own the product, but pays a regular fee.

Technical implementation

Next, we examine the technical dimension. We want to analyze how and with what effort the use case can be implemented and put this in relation to the economic potential. For this we discuss the following aspects, for example:

• Which sensors already exist, what should be installed in the future?
• Are there interfaces in the current hardware?
• What data, be it from the service organization, from error memories of the device or other sources exist and where are they currently stored?
• What is the deployment location?

Companies are often confronted with large amounts of data from various data sources when developing IoT platforms. This is where we provide support to find a suitable solution for storing this “Big Data”. There are numerous options for data management and data storage, depending on the format, collection frequency and other aspects of the data. Particularly well known are Data Warehouses and Data Lakes.

A particular challenge is to reliably collect deviations from normal operation. Here, Data Pipelines provide support by enabling continuous monitoring of data from initial data collection to long-term data storage. These also help to bring order to mostly unstructured data.

If relevant data does not yet exist, we assist with sensor selection and data collection planning. At all times, this is done with the use case in mind. With synthetic data, we can also accelerate implementation by simulating and testing scenarios early on.

Data analysis as an entry point

If data already exists, we perform exploratory analysis. The particular challenge here is to combine data from different sources. Simply put, all data sets need an intersection point so that they can be combined and viewed holistically.

After merging, we look for correlations to the unwanted situations. Found correlations are illustrated with the help of data visualizations and can be implemented in the long term with dashboards or automatic reporting.

In addition, we continuously check the quality of the data. Here we make suggestions on how data quality can be improved in the future through smart data processing.

The path to the finished product

In general, our Data Scientists take the approach of analyzing data to identify correlations and exploit them algorithmically. In this way, data-driven decision making is made possible. For this purpose, we use a variety of methods, frameworks and tools.

Depending on the application, simple statistical approaches can already lead to initial successes. These can be complemented by the development of individual algorithms. The development of machine learning models and other artificial intelligence approaches is also part of our portfolio.

grandcentrix offers many years of experience as an end-to-end IoT provider to accompany projects from a first test phase to productive deployment. We assist in moving from the problem to initial data collection, and building on that to implement value-added projects.