Comprehensive Environment Evaluation Model Of Engine Production Based On BP Mannual Neural Network

At present,the research object of the operational environment evaluation is generally a single factor.It does not consider the combination and antagonism between the various environmental factors,and the evaluation method is too subjective.In view of this problem,the comprehensive evaluation of engine production environment with multiple environmental factors was studied.First,the features,influences and hazards of the motor manufacturing environment was analyzed to establish the evaluation system of the environment,and grading standards of every single indicator were determined.And the three-layer BP neural network model was set up in sequence using temperature,humidity,air flow velocity,oil spray and noise as the input and the comfort level as the output.Bayesian regularization and momentum gradient descent method was introduced to solve the over fitting phenomena between high precision of training and low precision of forecasting in traditional BP neural network model.Experimental data indicates that the established model has a good performance in accordance with the reality and can instruct the improving of operation environment.Conclusions could be achieved from this dissertation:1.Using the data collected in several motor manufacturing plants in separated areas including noise,oil spray,temperature,humidity,air flow velocity,lightness and colors,and the impacts to operators;physical and mental health as the selection principle,an overall evaluation system including temperature,humidity,air flow velocity,oil spray,noise and lightness was established and used five levels(I,II,III,IV,V)to rank the environment.2.The hardware architecture of data acquisition process was constructed.Aiming at the index value with fluctuation characteristics,the Calman problem was transformed by the time series state space model,filter data was obtained.Then,the monitoring of sensor fluctuation was realized.For the abnormal index value,the monitoring data was corrected based on neighborhood nearest value mean filter.3.The comfort level model of motor manufacturing environment was set up based on BP neural network that can quantitatively evaluate the comfort level of environmental simulating traditional experts as well as eliminate the objectiveness and uncertainty of human evaluation.Bayesian regularization method was used to train and forecast for its performance.The trained network model proved its feasibility in the test with sample data.