Identification of acoustic emission sources in machinery; application to injection/combustion processes in diesel engines

The high temporal resolution of Acoustic Emission offers great promise in the on-line monitoring of complex machines such as diesel engines. The fuel injection process is one of the most important processes in the diesel engine and its timing and fuel delivery control are critical in combustion efficiency. In this work, the phenomena leading to the generation of acoustic emission during injection are investigated by simulation of the injection process in a specially designed rig and through test in running engines on a test-bed. Signal processing approaches are devised to produce diagnostic indicators for the quality of the injection process. The novelty of the research lies in; 1) obtaining a coherent set of data which allows the separation of the part of the signal associated with injection in a given cylinder from other sources adjacent in time and space, and 2) in developing a signal processing approach which allows this separation to be achieved on line using an array of sensors. As such, the research is generic to multi-source multi-sensor analysis in machines. A series of experiments were performed on an experimental injector rig, and two-stroke and four-stroke diesel engines under different operating conditions. The injector rig experiments provided useful information on the characteristic signatures of the injection events, finding which could be implemented to the more complex signal from the running engines. A number of sensor arrays (sets of two and three sensors) were used on two types of four-stroke engine at different running speeds to investigate the source identification of the injection events, the essential strategy being to add complexity to the information in the AE record by using engines of varying degrees of mechanical sophistication. It has been concluded that the AE signals are generated by the mechanical movements of the components in the pump and injector as well as aspects of the fuel flow through the injector and the piping. Also, it is found that the temporal structure of the AE is highly sensitive to sensor position, and that transmission path differences to a sensor array are generally large enough to allow source separation. Applying a purpose-designed thresholding technique, followed by canonical correlation allows the separate identification of parts of the AE signal in the short crank angle widow where sources involved in injection, inlet valve opening and combustion are operating