Real-Time Diagnostics for non-stationary Gear Noiseon Automotive Tooth Gears

摘要

The investigation of gear noise emitted by automotive tooth gears has gained increasingly in importance. These activities are predominantly focused on the reduction of acousticrnemissions and the detection of manufacturing gear failures. Commonly, standard procedures ofrnfrequency analysis and order tracking analysis based on fast Fourier transform techniques are applied.rnHowever, these signals are often characterized by non-stationary and non-linear properties which canrnhardly be investigated with ordinary spectral techniques. In many cases, the large amount of datarnproduced by sampling in audio quality prevents real-time analysis with conventional tools, which arerntherefore of only limited value for quality inspection. Therefore, a general concept for experimentalrnmeasurement (here realized for the field of vibrations) is presented. It has been implemented forrnseveral hardware families and is designed as a heterogeneous layer structure (combining andrnconnecting the features and performance of hardware, FPGA-, DSP- and PC-level programming) withrnMATLAB forming the core. A wavelet analysis treatment for streaming data has been implemented onrnDSP boards in real-time mode. The method was tested on one type of tooth failures and compared with wavelet analysis in post-processing mode and with conventional joint time frequency analysis. The objects under investigation were spur gears of balancer systems. These units are responsible for compensation of free inertia forces. It is demonstrated that non-stationary methods such as the proposed wavelet techniques are better suited for recognizing and in particular classifying manufacturing gear failures than the conventional FFT analysis techniques.