Indirect Adaptive Closed Loop Control of Solenoid Actuated Gas and Liquid Injection Valves

Solenoid actuated Injection valves are typically driven open loop by a pre-determined voltage or current profile. There is unit-to-unit variation of the valve electromagnetic/mechanical parameters as well as electrical supply transients, flow force transients and operating conditions while the drive voltage or current profile is fixed. Hence, by definition open loop drive is sub optimal. Extensive on engine calibration is necessary to correlate the movement of the armature to the current profile in the solenoid. Valve closure (the point at which the valve hits the stop corresponding to max armature stroke) is typically detected based on detecting an inflection point in the current profile. This method of closure detection is not reliable and is fundamentally flawed because the current profiles of valves can exhibit several other inflexion points due to bouncing (several closure events), non-linearity in valve inductance, back electromotive force (BEMF) characteristics, noise etc. In addition, the open loop approach requires a stiff power supply otherwise the valve closure might occur outside the narrow window used for closure detection.

We describe how we applied model based adaptive intelligence to reliably close the loop around valve closure using only the coil current feedback with the available CPU resources on the ECU without using special position sensors or other dedicated drive circuits. By appropriately controlling the drive voltage, we force the valve closure to correspond to a minimum (not an inflection) in the coil current feedback profile during the Pull-Hold event (See Figure). This phenomenon is due to the loss of Back EMF the first time the valve hits its stop. This was tested on a wide range of valves of different sizes and designs. Practically this means that we can now reliably detect and control the time at which the valve hits the stop to within the time resolution of the current feedback profile (currently +/-15 microseconds). Alternatively, we can control the value of coil current at closure. Large variations in bus voltage in a non-stiff power supply, flow force variations, aging etcetera can be tolerated due to the close loop nature. Now the only hard piece of information required from the application is the value of the current required to hold the valve against the stop to complete the injection cycle. The rest of the information, namely the main electromechanical parameters of the valve and the drive voltage waveform are automatically determined by the adaptive intelligence that is run between shots for each valve.

We will give actual measured final current profiles for the Woodward ERV (Electronic Rail Valve) and SOGAV (Solenoid Operated Gas Admission Valve) together with a video sequence showing how the coil current profile is adapted from shot to shot as the adaptive algorithm improves its estimates of the basic valve parameters while it keeps tracking the closure point. We will also discuss some of the practical hurdles that we encountered in developing this into a product. Testing a range of valves also yielded “interesting” new findings that are also reported.