Deep-Learning-Based Friction Modeling of Dry Interfaces for Structural Dampers

Friction-based dampers have gained attention as a cost-effective way to provide structural control during natural hazards. However, the dry friction interfaces in these systems result in a highly nonlinear damping response during the reversal of damper travel, termed damper backlash. Moreover, the stick-slip phenomena intrinsic to the sliding response of dry friction interfaces make the accurate modeling of friction-based structural dampers challenging. Dynamic friction modeling for structural dampers currently relies on analytical models to approximate the damper’s response at a current location given the damper’s state and average out the complex system responses during travel reversal or stick-slip movement to obtain a model of the system’s performance. In this chapter, we propose the use of a deep learning model to capture the temporal dynamics of the system that when combined with the LuGre friction model provides a physics-informed machine learning approach for inferring the damping force of a dry friction interface given the state of the model. Specifically, this chapter uses a long short-term memory model to infer the LuGre friction model’s parameters. A methodology for parameter identification using truncated backpropagation through time is given, which allows for real-time updating. Model validation is performed using a 9 kip rotary friction damper designed for high damping performance and mechanical simplicity. The model is validated with data from real natural hazard events and used in a real-time hybrid simulation. The performance, reliability, and accuracy of the deep-learning-based friction model are discussed.