As roll mills run in three-shift mode almost continuously, machine availability plays a hugely important role. Failures on a single machine ultimately delay the entire production process and thus cause extremely high costs. To prevent this, machines are normally maintained after a certain number of operating hours. On its new roll mill, KRAIBURG wanted to find out how to do this differently and took its first step toward the maintenance method of the future.
For some time, Hägglunds hydraulic radial piston motors from Bosch Rexroth have been driving the roll mills used to homogenize and roll out the rubber compounds. For its new machinery, KRAIBURG uses predictive maintenance services from Rexroth including the Online Diagnostics Network (ODiN). The core idea of this service package is to carry out maintenance work before a machine stands still by using a combination of sensors, cloud-based applications, and machine learning methods. In the new machinery at KRAIBURG, sensors first record detailed data about the oil tank, pumps, motors, and electric drive. The measurements taken include temperatures, oil levels, flow rates, and pressures.
The data collected is then sent to a Bosch Rexroth server, where it is analyzed using complex algorithms. The collected data, separately encrypted and handled for each customer, is transferred to the ODiN system. It is stored and processed on comprehensively secured Bosch servers in compliance with the Group’s very stringent data protection guidelines.
After installation of the new roll mill, ODiN initially collected data on all monitored components over a training phase lasting several months. Based on these signals, a machine learning algorithm determines a normal health condition for the roll mill. After the learning phase, ODiN uses a data-based model to continuously monitor the roll mill’s health index. If a single measured value moves outside the tolerance range for a short time, this does not necessarily lead to a—possibly unjustified—warning, as wear can rarely be detected with just one single signal. However, if the health index deteriorates due to changes in the data from several sensors, the system warns of a problem—even if the individual changes are within the defined limits. In the health index reports created on a regular basis, ODiN uses machine learning to provide appropriate information and helps to create specific recommendations for action.