Physics-Based Neural Network Models for Prediction of Cam-Follower Dynamics Beyond Nominal Operations

Abstract

Cam-follower mechanisms are key in various mechatronic applications to convert rotary to linear reciprocating motions. The dynamic behavior of these systems relies on the design parameters such as the cam shape and follower mass. It appears that for some combinations of system parameters, continuous contact between the cam and follower cannot be assured, leading to harmful periodic impacts. This research presents a data-driven approach to predict the influence of parameter settings on the system dynamics by learning from a limited data set of nominal operating conditions. More specifically, we present a hybrid model architecture encompassing an ordinary differential equation, consisting of a close interconnection of neural and physics-based network layers. Due to an increased generalization established by the physical laws, these physicsbased neural network models exhibit enhanced extrapolation capabilities compared to their black-box counterparts. Consequently, the presented models can accurately simulate the system behavior for parameter settings far beyond the nominal values included in the training data. This way, starting from a limited set of nominal time-series data, we could accurately estimate the set of critical system parameters that lead to hazardous jump phenomena in cam-follower systems.

Physics Informed Neural Networks for Control Oriented Thermal Modeling of Buildings

“…This work is similar to [27], where inductive biases based on the underlying physics laws are coded directly into the network. Several works have built upon this idea and have shown promising results in obtaining approximate solutions for difficult physics problems such as two body mechanics [28], heat transfer [29]. In [26,29], the encoded physics knowledge is strictly enforced on the predictions of the neural network and assumes the availability of complete physics.…”

“…In [26,29], the encoded physics knowledge is strictly enforced on the predictions of the neural network and assumes the availability of complete physics. Differing slightly from this approach, in [28], the authors enforce partially known physics and learn remaining physics parameters using the available data. These approaches show that trained models are better at extrapolating and require fewer training samples.…”

“…Consistent with previous works such as [26,28], we explicitly encode the underlying physics of the system in a standard neural network architecture and then train it on the data collected. Assuming a partially observable setting, the network is trained to predict the next observable state (x obs i+1 ) and a latent representation (z i ) using a high dimensional state input (x i ), action (u i ) and exogenous information (w i ).…”

Perspective: magnetic nanoparticles in theranostic applications

“…conservation principles) 134 , while another class directly employs physics-based layers in a DL model 135 or, alternatively, the DL model is used in compensating prediction discrepancies of the used physics-based model 136 . Grey-box approaches were shown to typically require significantly less training data compared to black-box models and could reveal unknown dynamics and system parameters 137,138 , allowing to improve currently existing models and e.g. find unknown nanoparticle dynamics.…”

Preventing Catastrophic Forgetting using Prior Transfer in Physics Informed Bayesian Neural Networks

“…This paper proposes a modelling methodology that exploits epistemic uncertainties to timely update the model and adapt to incoming operational data in a continuous fashion. We will rely upon prior hybrid physics-based data-driven models in which the unmodelled and unknown phenomena are taken up in a data-driven fashion and the well-known phenomena are described by underlying physical equations [38], [39]. These hybrid models are deterministic whereas here we use BNN as the data-driven component.…”