Press test bench for operation data and fault data generationCopyright: © ifas
Condition-based maintenance is a core element of modern maintenance strategies. The use of machine learning (ML) methods in particular, offers numerous advantages for the efficient implementation of such solutions. Furthermore, a comprehensive database covering a wide range of fault and operating scenarios is essential. On the hydraulic press test bench, topics relating to the generation, composition and processing of sensor and control data for condition monitoring are investigated.
Setup of the test benchCopyright: © ifas
The design of the test bench is based on the design of a hydraulic press. The press ram is coupled to a hydraulic cylinder, which is moved against a load in a position-controlled manner by using a proportional directional control valve. The hydraulic power is provided by a power unit and connected or disconnected to the working system by switching valves. A separate hydraulic load unit allows the flexible adjustment of different load profiles.
Reproduction of fault scenariosCopyright: © ifas
In addition to nominal fault-free operation, various faults and fault combinations of components can be continuously adjusted on the test bench. Bypass circuits at various points in the system allow adjustment of leakages, such as internal and external leakage on the cylinder of the working unit. In addition, hydraulic brakes along the guide rails of the ram can be used to introduce increased friction to the cylinder. Due to the modular design and easy accessibility, components can be quickly exchanged. This allows, for example, to install proportional directional control valves with different degrees of wear and to investigate their effect on the system behavior. The measurement equipment integrated in the test setup allows targeted adjustment and monitoring of targeted fault intensities at any time. Moreover, this enables to annotate the generated data records accordingly. Not only static fault states can be set, but also progressive degradation processes can be mapped according to the specification of a progression function.
Framework for fault classificationCopyright: © ifas
The generated data can be processed directly on the machine control unit or can be exported to external computing units for that purpose.
Based on a custom-developed framework, whose elements are in turn based on open-source code toolboxes, various sub-steps of the data processing chain can then be examined and analyzed with regard to an effective fault classification.
In the context of the press test bench, especially the blending or even the total replacement of data from the real machine with data from an associated lumped parameter simulation can be investigated. This can be done from the perspective of the composition of the database as well as the pre-processing steps, data characteristics and ML models used.