Research On Multi-vehicle Identification In Bwim Based On ICA

Bridge weigh-in-motion(BWIM)is an effective method to calculate vehicle axle weight even without touching the bridge structure.When a single vehicle passing the bridge,the traditional BWIM system can effectively identify the vehicle weight,especially the gross vehicle weight.However,BWIM shows no superoirity on the multiple vehicle presences due to the interplay of strain signals.In order to solve this problem,ICA is introduced to separate the mixed strain data under multiple presences circumstances to acquire the strain which is equivalent to the response caused by each vehicle.Comparing with other algorithms,like influence surface,embedded in modified BWIM,ICA is less dependent on the bridge information,and consequently more adaptable.This paper mainly studies the application of ICA to separation of multi-vehicle-induced strain signals through the FE analysis and some field tests on I-78 bridge,I-459 bridge,Alabama,United States and the northern approach bridge of Lunzhou Bridge,Qingyuan,Guangdong.The main research work of this thesis is as follows:(1)Based on the internal relation between the strains of each girder,an ICA model is built up and the major procedures to utilize ICA for identifying mult-vehicle is portrayed.(2)Three kinds of ICA algorithms,including WASOBI,FastICA and JADE,are compared.Then signal to interference ratio(SIR)and normalized cross correlation(NCC)are introduced as evaluation parameter and the evaluation results show that WASOBI is more stable and effective than the other two.Thus WASOBI is chosen to deal with the separation of real mixed signals.Besides,white noise is added to simulated signals for evaluating the anti-interference ability of this three ICA algorithms.(3)WASOBI has been proved effective through separating mixed signals obtained from field tests of three different types of bridges.An adjusted algorithm of axle weight calculation is proposed to identify axle weight through single-vehicle-induced strain signals,which are obtained from former separation results,The gross vehicle weight(GVW)and group of axle weight(GOA)are identified accurately while the results of axle weight calculation are not.According to the statistical characteristic of strain signals of each girder and corresponding separation results,some adjustments need to be done before applying ICA to separate real mixed strain signals of different types of bridges.(4)Two main influence factors of performance of signal separation: statistical independence and linearity of mixture model are taken into consideration.The results of factor analysis show that performance of FastICA and JADE both has a positive correlation with the statistical independence between original signals,while the performance of WASOBI is not seriously affected by the statistical independence.Although WASOBI algorithm is more adaptable than FastICA and JADE algorithm,FastICA and JADE algorithm perform better when the signal independence can be guaranteed.Thus,further analysis of the application range of these three ICA algorithms has been done.Moreover,the influence will be slight if the mixture model is high linearity,otherwise the separation result will be seriously affected.(5)Considering that ICA algorithms may fail when the original signals are correlated,non-negative matrix factorization(NMF)is introduced as a supplement.Based on nonnegative constraints,NMF can separate mixed signals without assumption of statistical independence.According to the good performance in separation of simulated signals and real signals,NMF is an alternative method for signal separation in BWIM system.