Success Story

A «Machine Learning» algorithm for the detection of abnormal operating behaviour

Assessment of friction-induced acoustic emission enables diagnosis of operating condition of sliding elements.

Tribological investigations of the friction and wear behaviour serve to simulate and assess the behaviour of mostly mechanical engineering systems or components, for example of components and assemblies in production plants or motor vehicles, mechanically and dynamically by means of suitable laboratory models. In doing so, the occurring wear mechanisms or friction conditions, which may lead to the failure of the component, can be studied under simplified, defined conditions and alternative material pairings and lubricants can be evaluated.

The failure of such components, for example journal bearings or guide rails, often occurs abruptly and without an obvious cause. Such a failure can destroy the affected components or the higher-level system and thus may lead to long downtimes and high costs for repair or maintenance. Equally, such failures often constitute a safety risk.

Based on a suitable analysis method, measures can be taken in time to prevent the destruction of the affected component. This was done by pattern recognition in measured sensor data, such as force, temperature, or acoustic emission, indicating imminent material or system failure. In addition to a variety of other sensors, as is usual for mechanical-dynamic investigations, a precision microphone was used for the project-specific task in order to record the airborne sound generated during the laboratory tests on self-lubricating ”sliding bushings” (sliding bronze bearings in contact with an axially oscillating steel shaft). The measurement signals recorded with the microphone were used to train a neural network-based machine learning (ML) algorithm ex-post, which distinguishes areas with normal from those with abnormal operating behaviour.

Figure: AC2T research GmbH

Impact and effects

With the developed ML algorithm, we are now able to detect abnormal operating behaviour with a high degree of accuracy on the basis of the acoustic emission measured during a tribological experiment. With regard to the simple sound measurement used, no changes have to be made to the system to be investigated. Since only a microphone with corres-ponding measurement electronics is required, this approach represents a simple and cost-effective method. In a further step, this method can be extended or adapted for online fault detection on systems in practical use.

Effective failure detection during operation makes it possible to shut down a system at risk in time to prevent major damages or to take preventive action – for example, the automated addition of lubricant.

In this way, damage to the affected components or aggregates can be kept to a minimum. This goes hand in hand with the potential to significantly reduce downtimes and repair costs as well as any safety risk in industrial plants.

Damage to a sliding bushing due to scuffing wear (Image: AC²T)

Project coordination(Story)

Dipl.-Ing. Dr. Josef Prost
Project leader
AC2T research GmbH

Partner