Data Acquisition

Data Acquisition (which is also called data logging) can provide a wealth of information about how a vehicle and/or driver is performing.

Back in 2001, I designed and built a data acquisition system for motorcycles.  I intended it mostly as a tool to evaluate my own riding.  The only commercial use I ever envisioned was for racing schools.  For those interested, I will provide some details on its workings in the section entitled On-Track.  But suffice it to say, the advent of commercially available GPS data loggers changed everything.  Most readers will find more value in the generally applicable information that I'll present next.  It's still relevant today.

Note to anyone considering a DA system: There will be a learning curve!  It's hard enough being your own mechanic while racing.  Tending the data acquisition system is almost a full-time job, and having a specialist helper is immensely beneficial.  Lacking a helper, you may be able to collect data while racing, but any analysis is probably going to be done after the fact.

Why Data Acquisition?

Eddie Lawson once said, speaking about an early DA system, “I hate that thing.  Here I think I'm at 100% throttle, but the computer says it's 25%.  I hate it!”  

An engine builder I know loved the idea of recording throttle position and said – only half joking – that he'd tell his customers, “Don't ask me for more horsepower until you're holding it wide-open all the time!”

In the notes I've kept on this project, there are three quotes from Paul Van Valkenburgh's book Race Car Engineering and Mechanics that guided me in the beginning:

“The best investment in cost per hour of learning time is probably an electronic performance recording system.”

“If a faster, more experienced driver is available, he can set a performance baseline to be matched.  Or two fairly good drivers can compare their recordings and learn from each other, since each will be better than the other in some distinct areas.  But even a single driver can learn by studying the output charts and seeing where his performance is not at the limit.”

“…it is more convenient to use a distance-base instead of a time-base so that all the laps come out exactly the same length on the chart paper and can be directly overlaid for comparison.”

As far as what needs to be recorded, in Drive to Win Carroll Smith complied his list for cars.  In order of priority:

• Engine RPM

• MPH

• Lateral Acceleration

• Steering Angle

• Longitudinal Acceleration

• Throttle Position

• Exhaust Gas Temperature

• Individual Wheel Speed

• Suspension Travel

• Ride Height 

My own order of priority (for a road racing motorcycle) is somewhat different.  The first group below is what's most important for rider development (but, of course, also aids vehicle development) – as that is where you'll get the most bang for the buck.  But generally, a superior rider on an inferior machine will almost always dominate an inferior rider on a superior machine.

• MPH.  By far the most useful data is the MPH record.  When graphed against track position, a conceptual map of the track emerges.  The peaks are the straightaways, and the troughs are the corners.  (Note, this was prior to the advent of GPS data-loggers.)

• Lap Time.  This should be self-explanatory.  Although this information will be available via Timing & Scoring, it may be difficult to correlate with other collected data.  

• Throttle Position.  Throttle control, on a road racing motorcycle, is everything.  And, as Eddie Lawson observed, that information is often surprising.  

• Engine RPM.  Although this can be back-calculated from MPH, gearing and gear-changes, it's simpler just to acquire it directly.  When MPH and engine RPM are known directly, it's obvious where gear changes have been made.

The next group is biased towards vehicle development, but things like suspension travel certainly can benefit rider development as well.

• Suspension Travel

• Exhaust Gas Temperature

• Cylinder Head Temperature

• Detonation Events

This last group are things I tried, learned from, and moved on.  They certainly are not mandatory.

• Airbox Pressure

• Air/Fuel Ratio 

• Track/Path Mapping (This is where commercially available GPS data-loggers changed everything.)

• Individual Wheel Speeds

• Power Valve Servomotor Position

• Power Jet Solenoid Actuation

Measuring Voltage

Some common specifications for an automotive DA system's voltage measurement capability would be a 0 to 5 volt range and a 12-bit resolution.  Automotive systems typically are configured for “unipolar” (only positive voltages) operation.  This alleviates the need to synthesize a power source having negative polarity.

A resolution of 12 bits means that a signal can be “quantized” into 1 part in 4096 – which may seem like a peculiar number, but it is simply 2 raised to the 12th power.  Similarly, an 8-bit resolution would provide 2⁸ (256) discrete levels and a 10-bit resolution (2¹⁰) would provide 1024 discrete levels.

However, more general-purpose DA systems can also accept “bipolar” (negative as well as positive) voltages.  Here a -10V to +10V range would be fairly typical, again with a 12-bit resolution.

When the input range (0 to 5V or ±10V) is considered, the smallest measurable voltage change can be determined.  This is simply the total range of voltages encountered divided by the number of quantizing levels available.  For example, a 0 to 5-volt range with a 12-bit resolution yields a step of about 1.2 mV (0.0012 V) between levels.

If the input range is extended to ±10V (while keeping the 12-bit resolution) the smallest quantifiable change would be 4 times larger, or 4.8 mV.

For comparison, a 0 to 5-volt range with an 8-bit resolution yields a smallest quantifiable change of about 19.5 mV (0.0195 V).  While high accuracy may be preferable, it is often unnecessary.  My DA system was configured for an 8-bit resolution.  Unnecessarily high resolution consumes memory more rapidly, which may result in fewer sample being stored. (Which was certainly a bigger concern back in 2001 than it is today.)

The other problem is that “system noise” is frequently the limiting factor in just how much resolution is actually usable.  You don't gain anything by being able to quantify a 1 mV change when successive measurements make the data appear to “bounce around” by 20 mV with no physical change in the thing being measured.

Similarly, the mechanical system you are measuring may itself set the limits.  If, for example, the “stiction” in forks only allows repeatable measurements to within, say, 1 mm – then being able to measure a 0.25 mm change is useless.

Sampling Rate

Sampling rate is simply the number of times per second a quantity is measured by the DA system.  It is specified in hertz (Hz).  1 Hz is once per second, 10 Hz is 10 times per second, and so on.

Automotive data acquisition systems typically sample signals that change relatively slowly.  It would be quite unusual to sample an automotive-type signal at 1 kHz (1000 times per second) — anti-lock brakes being an exception.  Sampling rates between 10 Hz and 100 Hz are fairly typical.  (However, sampling rates of many kHz, and even many MHz, are common in other types of data acquisition systems.)

The signal must be sampled sufficiently often so as not to “alias” it.  The Nyquist sampling theorem states that in order for information not to be lost, a signal must be sampled at least twice as fast as the highest frequency component present in that signal.  This minimum sampling speed is known as the Nyquist rate.  For example, you would have to sample a 60-Hz sine wave more than 120 times per second in order to mathematically determine its spectral content (i.e., to say it was a 60-Hz signal).  Note, however, that the Nyquist rate is nowhere near fast enough to make a picture of a given signal.  That is, sampling at the Nyquist rate, you could not tell if the sampled signal was a sine, square, triangular, or other wave shape.

From personal experience, I feel you need to sample at least one order of magnitude faster in order to make a decent picture.  So, for our 60-Hz sine wave, we would have to sample it at least 600 times per second — more frequently would make a better picture.  To see what I mean, draw a sine wave and then place 10 equally spaced dots along it.  This gives you an idea of how a data acquisition system can create a picture by interpolation.  And, just as with a digital camera —    the more pixels (samples), the clearer the picture.

Measuring Resistance

Many sensors exhibit a change in resistance that relates to a physical change.  Throttle position sensors, suspension pots, and thermistors are a some of the more common ones.

But resistance is not measured directly.  Rather, the sensing resistor becomes part of a voltage divider, and a voltage is ultimately what's measured.  Therefore, a known reference voltage (often 5 volts) is also required.  

Some devices (TPS and suspension pot) are themselves voltage dividers.  Other devices, like thermistors, also require a precision resistor to create the voltage divider. 

Thermocouples

An oversimplified, but useful, definition of a thermocouple is that a small voltage is produced when the junction of two dissimilar metals is heated.  This phenomenon is known as the Seebeck effect after the German physicist who discovered it.

Wikipedia lists 8 different types of thermocouples in common use.  They are referred to by letter (K, J, N, R, S, B, T, E).  But, by far, the most common thermocouple types in the automotive world are J, K, and T.  Type J is often used for CHT, type K is used almost exclusively for EGT, and type T is good for measuring coolant temperature.  Wires, jackets, and connectors are color-coded for identification.

Note that different countries use different standards for thermocouple wire colors.  This can cause no end of problems.  Here in the US, the standard is ANSI or ASTM and the negative lead is always red.  Of course, this is exactly the opposite of what most people think because the positive terminal of their vehicle's battery is red.  

Always verify the thermocouple standard / color code used in your country.

Note that a thermocouple exists (wanted or unwanted) wherever the junction of two dissimilar metals experiences a temperature different from that in the rest of an electrical circuit.  This highlights two interrelated factors critical to the successful use of thermocouples:

1. Temperature Difference

2. Dissimilar Metals

At some point in the thermocouple circuit, a connection must be made between the wires that form the thermocouple and the data acquisition system.  This point is known as the “cold junction” (because it is at ambient temperature, thermocouples typically measure hot things, and historically, the cold junction was an ice-water bath).  In order to accurately determine the temperature of the hot junction, the temperature of the cold junction must be known.  This is called “cold junction compensation,” and the DA system will have some means to accomplish this.

You can never avoid the transition from thermocouple wire to copper (on the circuit board, for example).  But you can control where it occurs.  DA systems that accept thermocouple inputs will have a means of measuring temperature very near the transition connection.  If there is a temperature gradient between where the DA system measures ambient temperature and where the thermocouple / copper interface occurs, an error will be introduced.  How significant that error is depends on the temperature difference.

I mention all this in order to explain why it's necessary to use the same type of metal wire all the way from the thermocouple back to the data acquisition system.  If you attempt to extend the thermocouple with plain copper wire, you would be, in effect, creating another thermocouple (remember: dissimilar metals experiencing a temperature difference).  This unintended thermocouple can either aid (add) or oppose (subtract) voltage to the circuit, thus introducing error.  The easiest solution is to start with long thermocouple wires and shorten them.

If you must lengthen the wires, then only “thermocouple extension wire” (the same type of metals as the measurement thermocouple itself) can be used.  The same goes for making a splice or at any connector interface – the proper metal types must be used, or an error will result.  So, what are the proper metals to use?

Thermocouple Materials

Different wire materials may be selected to optimize factors including cost, measurement range, linearity, and voltage swing.  With modern digital electronics, linearity is no longer a driving factor.

Some thermocouple materials are not elements.  They are compounds with the trade names Alumel, Chromel, and Constantan.  Alumel is composed of aluminum and nickel.  Chromel is composed of nickel and chromium.  Constantan is composed of copper and nickel.  

As you can see from the table below, the voltage swing over the entire measurement range is small indeed.  Take the type K for example: over it's intended measurement range of -270 to 1260 °C, it will only produce a total voltage change of 57.5 mV.  This is equivalent to an average change of 37.6 uV (microvolts) per degree C.  I emphasized the word average because the change is not linear, and requires a fairly high-order polynomial to produce a linearized result accurate within ± 0.7 °C.

Since the voltage produced is very small, it typically requires significant amplification (two orders of magnitude) prior to being sampled by the DA system.

A table for the most economical thermocouple types (no exotic metals) is listed below.  Different vendors may specify slightly different values, but the ones shown are representative.  Note that the higher the average °C per millivolt, the greater the temperature resolution for a given number of bits in the data conversion process.

 RTDs (Resistance Temperature Detectors)

High-temperature RTDs are expensive because they contain platinum.  The only place I have ever seen one used in an automotive application is a snowmobile's exhaust system.  Like a thermistor, an RTD's resistance changes with temperature, but it exhibits a positive temperature coefficient.  That is, resistance increases as the temperature of the sensor increases.  A resistance of 100 ohms at 0 °C is typical.

RTDs come in 2, 3, and 4-wire configurations.  They all require the same signal conditioning electronics — a precision current source and a way to measure voltage.  It's just how the three configurations get connected that differs.

There is no inherent difference in the response time for the signal conditioning electronics with an RTD versus a thermocouple.  It's the thermal mass of the sensor itself that dictates response time.  For RTDs, as with thermocouples, the faster the response the more delicate the sensor (and the more often you will have to replace it).