robotics//sensor//IMU//accelerometer

An accelerometer is an inertial sensor that measures the force per unit mass needed to carry a small proof mass along with its case, and it is used to find which way is down (the tilt of a drone or a phone), to sense motion and shocks, and to listen to the vibration of machines. In a MEMS accelerometer the proof mass hangs on silicon springs a fraction of a millimetre long; when the case accelerates the mass lags, the gap to fixed electrodes changes, and so does a capacitance that the chip turns into a number. A 16-bit part on a ±16 g range resolves about 0.49 mg per count; a wider range survives harder shocks and coarsens every step.

An accelerometer is an inertial sensor that measures the force per unit mass needed to carry a small proof mass along with its case, and it is used to find which way is down (the tilt of a drone or a phone), to sense motion and shocks, and to listen to the vibration of machines. In a MEMS accelerometer the proof mass hangs on silicon springs a fraction of a millimetre long; when the case accelerates the mass lags, the gap to fixed electrodes changes, and so does a capacitance that the chip turns into a number. A 16-bit part on a ±16 g range resolves about 0.49 mg per count; a wider range survives harder shocks and coarsens every step.

What it measures is specific force, acceleration minus gravity, f=a−gf = a - gf=a−g. Lying on a table it reads 1 g upward, although nothing moves; in free fall it reads zero, although everything does. The sensor is honest in an unintuitive way, and that one fact explains both its use and its limit.

Its use is tilt. At rest the reading is gravity alone, and the roll angle follows from two of its components, ϕ=atan2⁡(ay,az)\phi = \operatorname{atan2}(a_y, a_z)ϕ=atan2(ay​,az​): an attitude reference that never drifts, which is exactly what the gyroscope lacks.

Its limit is the tilt-acceleration ambiguity. The sensor cannot tell leaning from accelerating, so the tilt it gives is right only when the body is not accelerating much, and motor vibration rides on top of it. A drone therefore trusts the accelerometer for the long-term average and the gyroscope during manoeuvres (complementary filter).

Integrated twice it gives position, badly: a 10 mg bias is 180 m of error after a minute (dead reckoning). Its bias and scale are calibrated with the six-position test, each axis pointing up and down against ±1 g (sensor calibration).

In industry the vibration accelerometer is usually piezoelectric, bolted to a bearing housing with a stud so its response reaches tens of kilohertz, and its signal feeds vibration analysis at sampling rates of 10 to 25 kHz.