The aim of this project is to automate the maintenance process and to enable predictive maintenance. For this purpose, wireless sensor units, which supply energy from the machine heat and send sensor data to a central computer if required, are retrofitted. The combination of a retrofittable, energy-autonomous sensor platform and AI makes it possible to make production processes in the industry smart, efficient and reliable.
Especially the paper-producing industry faces great challenges in meeting quality demands and efficient production without unexpected downtime. The production of paper is very complex. At temperatures of up to 80 °C and a humidity of up to 100%, paper is produced in plants, that are several hundred meters long and consist of several hundred rolling bearings, drives and drive rollers. Each of these components can lead to a standstill and must therefore be regularly maintained. The components to be monitored, such as bearings and drives, are also found in other manufacturing industries such as the pharmaceutical and food industry. Nevertheless, the need for predictive maintenance in paper production is much higher, because these products, e.g. special papers for the furniture industry, play an important role and downtime leads to delivery delays and high follow-up costs.
The uniqueness of this project’s approach is the combination of wireless and energy-efficient hardware with intelligent algorithms to efficiently and reliably detect anomalies. The change detection on the microcontroller automatically learns at the gateway and detects all changes, these are transmitted by radio and thus the entire system is even more energy-efficient. Efficient and scalable predictive maintenance and OEE analysis can only be achieved in combination with energy-saving hardware and intelligent software.
In terms of data processing, the key idea and innovation is that the rather complex and compute-intensive algorithm for automatically detecting bearing state changes is divided into different parts. The various parts of the AI software run on different devices and have features that exactly match the corresponding hardware specifications. In this case, the endiio Retrofit Box and the endiio Gateway. The Retrofit Box algorithm reduces high-frequency sensor data (acceleration, vibration, gyroscope and magnetic field) to a few, particularly relevant features. The features must be condensed in such a way that the power consumption for the wireless transmission is minimal and at the same time they contain enough information for the higher-level algorithm to be able to detect changes in the state of the bearings. The algorithm on the gateway uses the previously extracted features from multiple states and automatically detects changes in state and behavior. The algorithm considers different aspects: time-dynamic behavior of each individual sensor data (i.e. acceleration) and their relation with each other, the state of each bearing and their relation with each other.