This paper presents artificial neural network (ANN) models which, considering physical variables (e.g. air temperature, relative humidity, horizontal luminance, sound pressure level, and density of people), qualitative and quantitative objective variables (e.g. gender, income level, age, education level and occupation), and subjective variables (e.g. behavior of respondent, brightness evaluation, acoustic comfort), are capable of predicting the subjective loudness in underground shopping streets. All models were properly trained for the prediction by determining the parameters for ANN, such as hidden layers, nodes, iterations, and training functions. Validation tests were conducted to ensure the accuracy of all the ANN models. Different ANN models have been developed, named general, group and individual models depending on their applicability. Prediction results indicate that the general model has the best generality, but its prediction correlation coefficient between outputs and desired targets is approximately 0.60. The individual model gives good predictions, with a correlation coefficient of approximately over 0.70 between outputs and desired targets. However, its applicability is limited. The group model has good generality, with a correlation coefficient of over 0.65 between outputs and desired targets. In conclusion, the group model is suitable for practical use in the design process. © 2012 Institute of Noise Control Engineering.

No abstract

Error compensation can improve the accuracy of machine tools effectively. Among the error sources affecting the accuracy of CNC machine tool, geometric error is always set as a key performance criterion. This paper summarizes several methods of geometric error modeling and reviews the characteristics of different methods. Furthermore, available methods for measuring geometric errors have been reviewed also based on the advanced instruments. This work aims at enhancing the efficiency of error detection and give a perspective for the application of error compensation in the future.

A review of methods used for fault recognition of the rolling elements in gearbox, namely gear and bearing, is presented in this paper. The procedure of the fault recognition can be classified into three phases, which are data acquisition, data processing or important features extraction, and the fault mode detection. Many different methods have been developed in dealing with each phase, so that the arrangement of the entire process has different strategies. This paper summarizes some general used and recently popular developed methods of each phase, and makes comparison between them. This work also aims at giving a perspective in such area in the future, based on the known researches.

No abstract