1.1 The Measurement Chain
about 11 minutes · 1 figure · quiz of 5 · Slides
After this section you can: name each link (mount, sensor, cable, conditioner, filter, converter, the controller's display) and say what each can do to the number.
Every number a test produces starts as motion at a point on a fixture or an article and ends as a figure on a screen or in a file. Between the two it passes through six links, and each one changes it. The mount carries the motion of the surface into the sensor. The sensor turns acceleration into an electrical quantity, a charge or a voltage. The cable carries that quantity to the rack, and on most sensors it also carries the power that runs them. The conditioner powers the sensor or converts its charge to a voltage, and applies a gain. The filter removes what lies outside the band the measurement is for. The converter turns the voltage into numbers. Then the controller or the recorder scales those numbers by a sensitivity somebody typed in, computes something from them, and shows the result in units and with a detector somebody chose. The screen is a seventh place for a number to go wrong, and it is the only place anybody looks.
Each link can lie in its own way, and the ways are worth knowing by name, because the rest of the chapter is organized around them.
The mount is a spring. The sensor's mass sits on it, and the pair has a resonance, the mounted resonance. Well below it the sensor moves with the surface and reports the surface's motion. Near it the sensor moves more than the surface and reads high. A loose mount has a lower resonance and rattles, and a rattle is broadband noise with spikes at the peaks of the motion, as the opening story showed. The mount is the subject of 1.4.
The sensor has a sensitivity, and the one the controller is using may not be its own. It responds a little to motion across its axis and, depending on how it is built, to bending in the surface under it and to changes in temperature. It has a range and a low-frequency limit, both fixed for most sensors on the day they were made. A sensor that has been dropped may have a different sensitivity from the one on its certificate. 1.2 and 1.3 take the sensor.
The cable has capacitance, which a long run turns into a low-pass filter on a powered sensor, and it can make its own noise when it moves. Its connectors work loose and let water in. Its shield is part of a ground path, and a ground path in the wrong place puts the line frequency into the measurement. The conditioner sets a gain and a low-frequency cutoff, and its gain is a second number that has to agree with the first. Both are in 1.5.
The filter and the converter are Chapter 2's. A filter removes what the sensor did outside the band, which is its job, and it can also hide a sensor that was overloaded by content outside the band, which is a trap (1.8). The converter has a range, and a signal beyond it is clipped; a signal far below it sits in the noise.
The display shows what the controller computed, never what happened. The sensitivity in the channel table (the controller's list of its inputs and their settings, 2.4), the units, the detector (peak, rms, which is the root-mean-square average, or peak to peak), the averaging and the frequency resolution all stand between the converter's numbers and the figure on the screen, and a wrong one of any of them changes the figure without changing anything about the motion.
The controller believes its inputs. It has no other way to know what the table is doing, and it holds whatever arrives on the control channel to the profile, the level the test asks for at each frequency, with the same confidence whether that signal is the motion or not. Chapter 3 puts it in one line, that the loop closes on the wire, and every link in this section is part of the wire. A wrong link is invisible on the control plot. A sensor reading twenty percent low gives a control plot in tolerance from one end of the band to the other while the article sees twenty percent more than it was asked for, and nothing on the screen is out of place.
Errors in the chain come in two kinds, and they are caught in two different ways. The first kind changes the number by a factor and leaves its shape alone: a wrong sensitivity, a gain entered twice, units in meters per second squared read as g, an rms detector read as peak. The spectrum has the right shape and the time history looks like a time history; only the scale is wrong. No amount of looking at the channel by itself will find it. It is found only by comparison with something independent: a known motion put into the sensor (1.7), a second chain at the same point, or the other channels at frequencies where everything on the table moves together (4.4's rigid-body check). The second kind changes the shape: a mount's resonance in the band, a high-pass that cuts the low end, a saturated amplifier, hum, a rattle. These show, to someone who looks at the right plot, and 1.8 is a field guide to them.
That is the reason for the second opinion: an independent channel beside the one that matters, recorded separately. Independent means it shares no link. It has its own sensor, its own mount, its own cable, its own conditioner, its own sensitivity entry, and where the stakes justify it its own recorder. Two chains that share a conditioner share its gain, and two channels that share a channel table share every typing mistake in it. When two independent chains at one point agree, every link in both has been checked against the other at once, for the cost of one more channel. When they disagree, the test stops until somebody knows why. 1.6 applies this to the control accelerometer, where it matters most, and Chapter 2 applies it to the recording.
Proof of a chain means a known motion in at the sensor and the expected number out on the screen, through every link the test will use: the sensor, on its cable, through its conditioner, on its channel, with its sensitivity, units and detector as the test will run. The hand-held calibrator in 1.7 is the tool for it, and 2.5 gives the full end-to-end check as a procedure, channel by channel, before the article is at risk. The order matters. The control channel is proven first, because it is the only channel the loop acts on, and an error there is not an error in the data. It is the test.
Decide. No channel is trusted until its chain has been proven end to end, from a known motion at the sensor to the expected number on the screen, through the cable, conditioner, channel and settings the test will use. The control channel is proven first. On any test the lab cannot afford to argue about later, a second chain that shares no link with the control channel sits beside it and is read at every level.
Common Questions
If the control plot sits in tolerance, has the control channel not proven itself?
No. The controller believes its inputs and holds whatever arrives on the control channel to the profile, so a sensor reading twenty percent low gives a plot in tolerance while the article sees twenty percent more (1.1). An error of that kind changes the number by a factor and leaves the shape alone. It is found only by comparison with something independent: a known motion at the sensor (1.7), a second chain at the same point (1.6), or the rigid-body check at the first low level (4.4).
Does a second sensor on a spare input of the same controller count as the second opinion?
Not as an independent one. Independent means it shares no link: its own sensor, mount, cable, conditioner and sensitivity entry, and where the stakes justify it its own recorder (1.1). Two channels in one channel table share every typing mistake in it, and two chains through one conditioner share its gain. A second channel on the controller still has a job as a watched backup (1.6).
Is a current calibration certificate the proof of a chain?
No. Proof is a known motion in at the sensor and the expected number out on the screen, through every link and setting the test will use (1.1). The certificate covers the sensor on the day it was calibrated, on the laboratory's mount, and says nothing about this cable, conditioner or channel (1.7). The hand-held calibrator is the tool, and 2.5 gives the full check.
Why count the display as a place for error, when it only shows numbers?
Because it shows what the controller computed with settings somebody chose: the sensitivity, the units, the detector, the averaging and the frequency resolution (1.1). A wrong one changes the figure without changing the motion. An rms reading taken for peak is off by 1.41, and meters per second squared read as g by 9.81 (1.7).
Quiz: Try It
Open book. 5 questions drawn from this section's pool. 4 correct passes the quiz. Nothing is recorded here; in the course your passing attempt opens the next section.