Prevention made simple

Not applicable

Predictive maintenance reduces system downtimes and thus saves time and money

Unexpected machine downtimes are a nightmare for every production manager. Particularly in large plants that produce around the clock, such as mines, foundries, or rubber compound plants, every hour a machine is not running is linked to extremely high costs. Forward-looking maintenance is a reliable method of reducing expensive plant downtimes in practice.

An Online Diagnostics Network (ODiN) service package uses the interaction of sensors, cloudbased applications, and machine learning methods to monitor operating conditions based on models and to carry out predictive maintenance activities. During the teach-in period, a machine learning algorithm identifi es a normal healthy condition from a variety of sensor signals, for example pressure, flow, vibration, temperature, and oil quality, according to the assembly to be monitored.

After the teach-in phase, the online system uses the data-based model to continuously calculate a health index for the monitored assembly. If a single measured value moves outside the tolerance range for a short time, this does not necessarily lead to a (false) warning, as wear can rarely be detected with just one signal. If the health index deteriorates due to changes in the data from multiple sensors within the defi ned limits because the behavior of the machine has changed, the system warns of a problem. The health index generated not only indicates the status of the actual assembly being monitored, but also more gradual changes in the upstream and downstream mechanical or hydraulic systems. If movements start to take longer or require more power over time, this is a sign of wear in the mechanical or hydraulic system.

An example data record shows how much of a complex task wear diagnosis can be: From a statistical perspective, the probability of a fault being randomly discovered is just 13 percent. An expert who uses traditional methods to continuously monitor the system will have a 43 percent probability of detecting it. However, the system has a fault detection rate of over 95 percent. The software can either precisely identify the fault, or it supports maintenance engineers with troubleshooting by localizing the fault to a particular assembly. The components that interact to carry out predictive maintenance are shown in the infographic.


Must Read


Related Topics