Tweet
In response to the posting by Jon Fewster, we offer the following
information on the use of accelerometers for human body motion analysis. Our
experience has been mainly with micro-machined silicon piezoresistive
accelerometers. We do not recommend using silicon accelerometers made by
older methods (e.g.: Entran and others) because of high cost, fragile wiring
and change in electrical characteristics after impact loading. Suppliers of
micromachined devices include IC Sensors, NovaSensor and SenSym. (SenSym, I
have been told, has been bought and may leave the field.) Each uses
slightly different chip design and packaging; none seem interested in making
a 3-axis package.
Note also that piezoresistive sensors detect the steady-state pull of
gravity as well as inertial acceleration due to the start and stop of
motion. Piezoelectric sensors can detect only the latter, with a
low-frequency cutoff typically about 10 Hz, causing slow motion data to be
lost. The gravitational component superimposed on the inertial component
prevents simple integration to calculate change in position, but has the
advantage of providing a continuous vertical reference. Integration is also
made difficult by a 1 to 2% error due to gravity in the non-sensitive axis
direction. These constraints led us to decide upon pattern-matching,
peak-detection and Fourier spectrum analytical methods, rather than try to
derive absolute position by double integration. We also use the sensors as
tilt-meters, by filtering out signals above about 0.2 Hz, on the assumption
that higher frequencies represent the inertial component (valid only for
constrained motions). If the signal is high-pass-filtered, these
accelerometers are equivalent to piezoelectric ones, but without need for
charge amplifiers.
We are using 15 G range IC Sensors (Milpitas, CA, 408-432-1800) devices,
most recently in the model 3031 surface mount package. At 0.3 X 0.3 X 0.14
inches, these are small enough for unobtrusive mounting on the body, even
after assembly of three sensors for multi-axis sensing. Cost is very
reasonable ($58/channel) compared to piezoelectric devices. The company
supplies a calibration sheet for each sensor only at extra cost; since there
are significant variations in offset and slope of the voltage- to-
acceleration curve within a batch of sensors; we calibrate each sensor
periodically in a fixture that places it with its sensitive axis up, down
and horizontal (11 and 0 G). Two out of about 40 sensors have failed
(showing uncontrolled drift, probably due to oxide in an internal
connection) and were replaced by the company under warranty. The company
has a long lead time (90 days) before delivery of 15 G 3031 accelerometers,
due to selection for high sensitivity out of batches of 120 G sensors; the
company plans to produce batches specifically for 15 G range as demand
increases and techniques improve.
Presently, the cables from the sensors are plugged into a custom-made
amplifier board inside a self-contained wearable computer/recorder. This
allows the wearer to move about without being tethered to a fixed computer.
The cost and complexity of a radio telemetry system for 12 data channels,
and the need to acquire data outside the laboratory, precluded transmitting
the sensor outputs to a fixed receiver. After 40 to 180 minutes of data
collection, a telephone-type modular cable is plugged into the wearable
recorder to send the stored data to a fixed computer.
While the wearable recorder's microcomputer is slow and thus cannot perform
the real-time signal analysis we eventually want to do, it functions quite
well for taking data at up to 50 samples per second with 12 channels per
sample accurate to one part in 1024. Recording is started, stopped and
coded for identification of the wearer's activity using an infrared remote
control, so the experimenter does not have to have physical contact.
The need for 12 channels arises because we are trying to describe both
translational and rotational motion of the head and trunk, requiring a
minimum of 6 degrees-of-freedom assuming each is a rigid body (not exactly
true, but close enough). Fewer channels causes ambiguity and a high number
of false alarms if the system is used as a fall detector. We anticipate
clinical use of accelerometric instrumentation to occur in three stages:
first, as a diagnostic tool to quantify hitherto qualitative measures of
balance; second, as a biofeedback device during therapy; and third, as a
fall-prevention aid P which might be called a 'balance orthosis' P for
continuous use by institutionalized and community-living fall-prone elderly.
The VA Rehabilitation R&D Center encourages establishing multi-center
collaborations for evaluation of devices we have developed, and a number of
such collaborations are explicitly included in the recently approved 3-year
extension of the grant under which this work is being carried out. We are
open to additional collaborations to the extent of our resources. We are
also open to negotiation of Cooperative R&D Agreements with private industry
for commercialization of this technology. If you are interested in pursuing
either of these options, please contact me (voice: 415-493-5000 x 3345).
Eric E. Sabelman, PhD
Section Chief, Human/Machine Integration
Palo Alto VA Rehabilitation R&D Center
information on the use of accelerometers for human body motion analysis. Our
experience has been mainly with micro-machined silicon piezoresistive
accelerometers. We do not recommend using silicon accelerometers made by
older methods (e.g.: Entran and others) because of high cost, fragile wiring
and change in electrical characteristics after impact loading. Suppliers of
micromachined devices include IC Sensors, NovaSensor and SenSym. (SenSym, I
have been told, has been bought and may leave the field.) Each uses
slightly different chip design and packaging; none seem interested in making
a 3-axis package.
Note also that piezoresistive sensors detect the steady-state pull of
gravity as well as inertial acceleration due to the start and stop of
motion. Piezoelectric sensors can detect only the latter, with a
low-frequency cutoff typically about 10 Hz, causing slow motion data to be
lost. The gravitational component superimposed on the inertial component
prevents simple integration to calculate change in position, but has the
advantage of providing a continuous vertical reference. Integration is also
made difficult by a 1 to 2% error due to gravity in the non-sensitive axis
direction. These constraints led us to decide upon pattern-matching,
peak-detection and Fourier spectrum analytical methods, rather than try to
derive absolute position by double integration. We also use the sensors as
tilt-meters, by filtering out signals above about 0.2 Hz, on the assumption
that higher frequencies represent the inertial component (valid only for
constrained motions). If the signal is high-pass-filtered, these
accelerometers are equivalent to piezoelectric ones, but without need for
charge amplifiers.
We are using 15 G range IC Sensors (Milpitas, CA, 408-432-1800) devices,
most recently in the model 3031 surface mount package. At 0.3 X 0.3 X 0.14
inches, these are small enough for unobtrusive mounting on the body, even
after assembly of three sensors for multi-axis sensing. Cost is very
reasonable ($58/channel) compared to piezoelectric devices. The company
supplies a calibration sheet for each sensor only at extra cost; since there
are significant variations in offset and slope of the voltage- to-
acceleration curve within a batch of sensors; we calibrate each sensor
periodically in a fixture that places it with its sensitive axis up, down
and horizontal (11 and 0 G). Two out of about 40 sensors have failed
(showing uncontrolled drift, probably due to oxide in an internal
connection) and were replaced by the company under warranty. The company
has a long lead time (90 days) before delivery of 15 G 3031 accelerometers,
due to selection for high sensitivity out of batches of 120 G sensors; the
company plans to produce batches specifically for 15 G range as demand
increases and techniques improve.
Presently, the cables from the sensors are plugged into a custom-made
amplifier board inside a self-contained wearable computer/recorder. This
allows the wearer to move about without being tethered to a fixed computer.
The cost and complexity of a radio telemetry system for 12 data channels,
and the need to acquire data outside the laboratory, precluded transmitting
the sensor outputs to a fixed receiver. After 40 to 180 minutes of data
collection, a telephone-type modular cable is plugged into the wearable
recorder to send the stored data to a fixed computer.
While the wearable recorder's microcomputer is slow and thus cannot perform
the real-time signal analysis we eventually want to do, it functions quite
well for taking data at up to 50 samples per second with 12 channels per
sample accurate to one part in 1024. Recording is started, stopped and
coded for identification of the wearer's activity using an infrared remote
control, so the experimenter does not have to have physical contact.
The need for 12 channels arises because we are trying to describe both
translational and rotational motion of the head and trunk, requiring a
minimum of 6 degrees-of-freedom assuming each is a rigid body (not exactly
true, but close enough). Fewer channels causes ambiguity and a high number
of false alarms if the system is used as a fall detector. We anticipate
clinical use of accelerometric instrumentation to occur in three stages:
first, as a diagnostic tool to quantify hitherto qualitative measures of
balance; second, as a biofeedback device during therapy; and third, as a
fall-prevention aid P which might be called a 'balance orthosis' P for
continuous use by institutionalized and community-living fall-prone elderly.
The VA Rehabilitation R&D Center encourages establishing multi-center
collaborations for evaluation of devices we have developed, and a number of
such collaborations are explicitly included in the recently approved 3-year
extension of the grant under which this work is being carried out. We are
open to additional collaborations to the extent of our resources. We are
also open to negotiation of Cooperative R&D Agreements with private industry
for commercialization of this technology. If you are interested in pursuing
either of these options, please contact me (voice: 415-493-5000 x 3345).
Eric E. Sabelman, PhD
Section Chief, Human/Machine Integration
Palo Alto VA Rehabilitation R&D Center