Regression Model for a Sensor‐Integrating Jaw Coupling—Compensation for the Viscoelastic Material Behavior

ABSTRACT

In recent years, there has been an increasing demand to digitize production processes. In order to obtain the necessary data for these processes, numerous sensors must be installed on or in the machines. This can be difficult, especially when done retroactively, as the existing installation space must be able to accommodate the sensors. To address this issue, sensor‐integrating machine elements are currently being developed. In these machine elements, the sensors are integrated directly into the respective element, and thus enabling a space‐neutral installation. In the current work a sensor‐integrating jaw coupling is investigated. The aim of this coupling is to obtain the transmitted torque via measuring the deformation of the gear rim teeth. For this purpose, dielectric elastomer sensors (DESs), which change their capacitance when deformed, are employed. The functionality of this principle was shown in previous works. The gear rim material has viscoelastic properties. Thus, hysteresis occurs when plotting the torque (the measurand) versus the capacitance (the measured quantity). It follows, that to predict the torque based on the measured capacitance, a regression model is needed. For this task, two approaches are compared with each other: (i) a Gated Recurrent Unit (GRU), a type of recurrent neural network, and (ii) a physics‐based approach using a combination of analytical models with a GRU. The hyperparameters of the models were determined using the Python library “Optuna.” The models were trained using numerical data of the sensor‐integrating jaw coupling. The two models are compared in terms of the computational time required and their level of accuracy.