Classification of interior noise comfort level of Proton model cars using artificial neural network

• Main Author: Allan Melvin, Andrew
• Format: Thesis
• Language: English
• Published: Universiti Malaysia Perlis (UniMAP) 2014
• Subjects: Vehicle Noise Comfort Index (VNCI); Interior noise comfort; Artificial neural network; Car interior noise; Automobiles > Component and parts; Noise control; Automobiles > Interior acoustic
• Online Access: http://dspace.unimap.edu.my:80/dspace/handle/123456789/31255
• Institution: Universiti Malaysia Perlis

Car interior noise comfort level classification is one of the most promising sub-fields in automotive research. Car interior noise comfort indicator is developed to help the drivers to keep track of the noise comfort level in the car. Determination of car comfort is important because continuous exposure to the noise and vibration leads to health problems for the driver and passengers. In this research, a proton model cars noise comfort level classification system has been developed to detect the noise comfort level in cars using artificial neural network. This research focuses on developing a database consisting of car sound samples measured from different proton make cars in stationary and moving state. In the stationary condition, the sound pressure level is measured at 1300 RPM, 2000 RPM and 3000 RPM while in moving condition, the sound is recorded while the car is moving at constant speed from 30 km/h up to 110 km/h. dB Solo equipment is used to measure the noise level inside the car. Subjective test is conducted to find the jury’s evaluation for the specific sound sample. The data is preprocessed and features are extracted from the signal frames. The correlation between the subjective and the objective evaluation is also tested. The feature set is then feed to the neural network model to classify the comfort level. The respective index is displayed at the designed Graphical User Interface (GUI). Experimental results show that the use of proposed Composite Feature yields a better classification accuracy compared to the conventional feature extraction method. The Spectral Composite Feature gives the highest classification accuracy of 94.21%.