Sensors

Select a physical response quantity at a location, then apply an ordered and reproducible measurement model without altering the structural state.

Sensor types

Sensor	Source history	Quantity value
DisplacementSensor	result.displacement	"displacement"
VelocitySensor	result.velocity	"velocity"
Accelerometer	result.acceleration	"acceleration"

from dynaresp import sensors

layout = [
    sensors.DisplacementSensor(position=0),
    sensors.VelocitySensor(position=1),
    sensors.Accelerometer(position=2),
]

For MDOF systems, position is an integer DOF index. For beams, it is a physical coordinate along the span.

Measurement pipeline

flowchart LR
    R[State history] --> O[Observation vector]
    O --> C[Clean signal]
    C --> E1[Sensitivity]
    E1 --> E2[Bias]
    E2 --> E3[Noise]
    E3 --> E4[Clipping / quantization / dropout]
    E4 --> M[Measured signal]

Effects run in the order provided:

accelerometer = sensors.Accelerometer(
    position=2,
    effects=[
        sensors.Sensitivity(1.02),
        sensors.Bias(0.05),
        sensors.GaussianNoise(snr_db=30.0),
        sensors.Clipping(-20.0, 20.0),
    ],
)

The noise= argument is a convenience that appends one effect after the explicit effects sequence:

accelerometer = sensors.Accelerometer(
    position=2,
    noise=sensors.GaussianNoise(standard_deviation=0.01),
)

Available effects

Effect	Parameters	Operation
GaussianNoise	snr_db or standard_deviation	Adds independent normal noise
Bias	value	Adds a constant offset
Sensitivity	factor	Multiplies the signal
Clipping	minimum, maximum	Limits the measurement range
Quantization	bits, full_scale	Uniformly quantizes a symmetric range
Dropout	probability, value	Replaces randomly selected samples

GaussianNoise requires exactly one of snr_db and standard_deviation. SNR-based noise computes the RMS of that sensor's clean signal. A zero clean signal therefore receives zero SNR-based noise.

Clean versus measured output

import numpy as np
import dynaresp as dr

spec = dr.SimulationSpec(
    system=system,
    load=load,
    sensors=(accelerometer,),
    time=np.arange(500) / 100.0,
    positions=np.arange(system.ndof),
)
result = dr.run_simulation(spec, seed=42)

clean = result.clean_signal[0]
measured = result.signal[0]

The structural state histories and clean_signal are never modified by sensor effects. This makes the result suitable for denoising, calibration, and robust learning experiments with exact ground truth.

Randomness

Every sensor receives an independent random generator derived from the simulation generator. Each effect receives a child generator in turn. A fixed simulation seed reproduces the complete measurement chain.