Description: A fully synthetic time-series dataset generated from a physics-based simulation of a 2D radial rubber bushing under displacement-controlled excitation. The dataset is designed for training, validating, and benchmarking real-time anomaly detection models intended for deployment on resource-constrained hardware (FPGA).The mechanical model is a nonlinear viscoelastic system with Coulomb friction (two Maxwell branches). Prescribed displacement profiles (sine, triangle, ramp) are applied and the resulting reaction force and internal states are computed at 1 kHz. Realistic sensor effects are applied to the three observable channels: additive Gaussian noise and optional first-order low-pass filtering on displacement and force. Structure: The dataset contains 90,000,000 timestep rows from 100 bushing families × 30 simulation runs (3,000 runs total), each 30 seconds at 1 kHz. Column Type Description family_id int Bushing parameter family (0–99) simulation_id int Run index within family (0–29) t float Time in seconds x, v, F float Ground-truth displacement, velocity, reaction force z1, z2 float Maxwell branch internal states (ground truth) x_meas, v_meas, F_meas float Observable sensor channels (model inputs) anomaly int 1 if anomaly active at this timestep, else 0 anomaly_type str Active anomaly type or normal Five fault modes are represented: stiffness softening step (param_step), friction increase ramp (param_ramp), force spike (signal_spike), force dropout (signal_dropout), and combinations thereof. The normal class comprises approximately 81.6% of timesteps. Model inputs are restricted to x_meas, v_meas, and F_meas (3 channels). The recommended input window is 512 samples (stride 256). Ground-truth columns (x, v, F, z1, z2) are included for analysis only and must not be used as model inputs. Splits should be performed at the simulation-run level to prevent data leakage.