ANFIS modeling of vibration transmissibility of a power tiller to operator

This research was conducted to evaluate a 13-hp power tiller vibration transmitted to operator's body experimentally at three-axis on transportation conditions. Also, to survey the usefulness of ANFIS in predicting vibration transmissibility (TR) from seat (WBV) and handle (HTV). Vibration acceleration signals were measured at the power tiller handle, trailer seat, and the operator's wrist, arm, chest and head positions for four transmission gear ratios, five engine speeds, and three directions of measurement. The results indicated that the TR value was the greatest at the operator's wrist, then at the operator's arm, chest and head, respectively for all the gear ratios and engine speeds. The TR value at different body organs was dependent on measurement direction. The maximum TR was observed at the operator's wrist and arm on the longitudinal direction and at the operator's chest and head on the vertical direction. The frequency domain results showed that TR from the trailer seat to the operator's chest was more than 100% in frequency range of 1-16 Hz. The TR was above 100% for the handle to the operator's wrist in the frequency range of 3.15-6.3 Hz and for the operator's wrist to the arm from 4 to 5 Hz. The TR more than 100% was observed at frequency range of 1.25-2 Hz for the operator's chest to head. The TR from the operator's wrist to the arm and the operator's arm to the chest was not more than 100% in any frequency bands. The results also revealed that the TR from the seat to the operator's chest was more than those of the handle to the arm and the handle to the wrist for all the three directions of vertical, lateral and longitudinal. The ANFIS methodology successfully learned the relationship between the input and output data sets. It can be concluded that the performance of subtractive clustering method was slightly better for HTV-TR(t) modeling (total TR) than grid partition method. In general, the performance of the ANFIS modeling for HTV-TR(t) is slightly better than for HTV-TR modeling, and this trend was also observed for WBV modeling. Because human response to multi-axis vibrations could not be easily understood using available laboratory equipment, investigation of human response to these vibrations in real conditions and modeling the results using machine learning and AI systems could increase our understanding of the biodynamic responses of the human body to multiple-axis excitations and assist the development of biodynamic models of the body.