Engine Testing and Dynamometers

There's lots of information on 'power' dynamometers on the internet. But that's not what the OEM's really work with, and not what I'm going to randomly write about today. The 'Big Boys' use a dynamometer that's quite old in basic design, yet quite sophisticated at the same time. A 'real' dynamometer is made of a simple AC induction motor and a fast-responding variable frequency drive, and a way to measure torque. Normally, an 'inline' torque transducer is used - Himmelstein, Magtrol, and Honeywell make suitable units. The VFD can be outfitted with braking resistors to dissipate the power, or the power may be regenerated and fed back into the electrical grid. Some arrangements include a more traditional waterbrake or eddy current dynamometer in tandem with the AC unit; this is used to control the cost of the system. Normally, a motoring dynamometer needs to be sized roughly 3:1 - for every 3kW of absorption, you need about 1kW of motoring. This ratio, along with reasonable inertia management, should allow the control system to follow all known engine emissions traces. Care must be taken to bias and control the gains of the absorption dynamometer and AC motor in the tandem arrangement so that instability is avoided.

Measurement of emissions is not normally done with those little handheld 'sniffers' that your local garage has. Large racks by Horiba and AVL are carefully designed for proper emissions sampling, and are far more sensitive and responsive than the typical mechanics tool. Some test labs (notably, SWRI and Electronic Microsystems) purchase analyzers and integrate them into racks themselves. Other companies purchase ready-made and validated emissions test racks. Modern test cycles require careful time synchronization between the engine, dynamometer, and emissions sampling.

A 'real' emissions test cycle is a fairly long test - not the couple minutes of a roadside sniffer test. A US heavy-duty diesel transient test takes about an hour to do - a 20 minute test, a short heat soak, followed by a repeat of the 20 minute test. A couple of hours after that you get your four number result. The bagged samples need to be processed and the hydrocarbons need to be analyzed on a gas chromatograph to split the various species apart. The WHTC (world harmonized transient cycle) takes just over an hour. That one is a 30 minute test, a five minute heat soak, and again, a repeat of the 30 minute test. The two test results are averaged; the cold test is weighted at around 1/6 of the test result with the hot cycle making up the other 5/6. The engine starts out at 'room temperature' - 25 degrees C, and within a few degrees. Both test cycles consist of 1 Hz sampled torque and speed data, with the assumption that linear interpolation be applied between points. The US HDDT cycle requires the use of a CVS and bagged samples, whereas the WHTC allows either approach.

From the emission test management and data acquisition system, measured and setpoint data are compared using linear regression. Intake air temperature and humidity are monitored to ensure that they remain within tolerance. If all of this data matches and is within prescribed limits, the emission test is considered valid. For engine speed, typical limits are: 0.95 < m < 1.03; -50 < b < 50; r2 > 0.97. These are relatively tight limits but for any competently designed control loop on an AC dynamometer, is actually quite trivial. For engine torque, which is much harder to measure and control accurately, typical limits are 0.83 < m < 1.03; -20 < b < 20; r2 > 0.88. Engine speed is primarily controlled by a the AC dynamometer, which has a very fast control loop response. Normally, a 50 Hz (or faster) bandwidth is available and with relatively insignificant delays. This allows fairly high PID gains to be used. The torque loop consists of the engine dynamics, engine rotational inertia, and engine fuelling and airflow delays. A spark ignition engine can be quite challenging to control because of the significant delay from pedal movement to torque generation (engine filling delay of an average of 0.75 crankshaft revolutions), and by the low-pass filtering nature of the intake manifold volume. Relatively low torque PID gains must be used, and a feedforward approach to engine torque control must be used instead. Normally, a 'pedal map' is generated based on steady-state mapping of the engine - this is strictly for the emissions test. Alternatively, if the accelerator pedal is linked more-or-less directly to a throttle, or has a constant-power control law, a simple calculation based on desired torque and measured speed will produce a throttle position that is very close to the desired value. Relatively low PID gains can then correct the throttle position to achieve the desired engine torque setpoint. Many diesel engines are easier to control in terms of torque response. A notable exception is the Cummins N14 which uses a fill-wait-pump cycle.

Emissions data can be gathered and digitally integrated to generate emissions test numbers. Diluted or raw exhaust gas may be used, but the trend has been towards using raw exhaust gas measurement because the emissions standards are getting that much tighter, and control of dilution air contaminants is not a trivial matter. Thermal management is important in raw sampling. Condensation is an issue and will result in inaccurate emissions readings and unpredictable time skews. De-humidifying the sample is an option, however, emission values will be affected because some pollutants are water-soluble. Either heated analyzers (HFID, HCLD) or correction factors must be used. Obviously, emission test results with correction factors are not as accurate as those directly measured. Some data acquisition systems show not only instantaneous PPM readings, but also instantaneous g/kWh and up-to-now g/kWh averages, which are very valuable for engine calibration. Some data acquisition and control software packages allow for replay of only part of the emission test cycle for calibration purposes.

A pet peeve... people who think that engine control is basically a 'classical' feedback control system that you just put a couple of PID controllers on. It is not. Control complexity is present because the time lag of the 'plant' is long compared to the required response time. Much of what goes into control of an engine is feed-forward, open-loop, achieved by table lookups. There are feedback loops and observers and predictors, but engine calibration still consists of data gathering, very much like engine calibration of a few decades ago. Sure, we now use model-based controls which are more accurate and easier to quantify. Observers and self-tuning regulators and H-infinity controllers are also used, but the key is that the open-loop response is the critical thing. Another interesting tidbit... I was looking at the Apollo Guidance Computer design. I could do an engine controller with one, although the timer resolution (6400 Hz) is a bit spotty for accurate fuel injector and ignition timing. It would run well, but probably not be quite as controllable from an emissions standpoint as I'd like. Most modern microcontrollers are very much based on the architecture and layout of the Apollo Guidance Computer.