Study On Variation Pattern Recognition Based On LSTM For Biw Ocmm Data

The assembly quality of Body-In-White(BIW)directly affects the performance of the automobile such as the sealing,wind noise,and smoothness.How to improve the quality of BIW continuously has always been the focus of major auto manufacturers.With the gradual promotion and application of Optical Coordinate Measuring Machine(OCMM),a large amount of manufacturing process inspection data has been generated.Identifying the data deviation model accurately is conducive to fault location and diagnosis rapidly.However,the random disturbance of welding process is frequent and the production environment is open.OCMM data includes the influence of multi-source noise,which makes it difficult to extract and use information effectively.Moreover,the existing pattern recognition methods rely on subjective experience for feature selection,and ignore the time dimension parameters,which results in low recognition accuracy.Aiming at the above problems,In this paper,a deep learning long short-term memory(LSTM)neural network is introduced to monitor the running status of the BIW production line and identify the anomaly pattern of the welding process automatically and intelligently.The main content of this paper can be divided into three parts:(1)OCMM accuracy analysis and denoisingAn anomaly deviation pattern separation method based on wavelet denoising theory is proposed to assess the standard deviation of OCMM data accurately.The Jarque-Bera method is used to test the normality of the fluctuations of different layers,and the fluctuations closest to the normal distribution are selected as the noise of the manufacturing system of BIW,avoiding the subjective selection for estimation of the threshold.It provides support for subsequent pattern identification of anomaly deviations in OCMM data.(2)Deviation pattern recognition based on LSTMThe feature extraction method for OCMM data is relatively single.The LSTM network suitable for processing time-series information is introduced.The model of anomaly deviation recognition is constructed by adaptively extracting features,and experiments are conducted on Keras open source platform.Compared with the pattern recognition method of BP neural network based on feature fusion,it is proved that the algorithm has obvious advantages in the recognition ability and generalization ability.(3)On-line intelligent monitoring of BIW welding qualityOCMM data analysis and processing based on wavelet theory and the anomaly deviation pattern recognition method based on LSTM are integrated,and the on-line monitoring framework of BIW quality is put forward.The proposed algorithm is verified by two typical examples of deviation control of BIW.In summary,the LSTM neural network is introduced into the analysis and processing of OCMM data,studies of dynamic pattern recognition and on-line quality monitoring.The method illustrated in this paper can not only be used to offer new theoretical orientation and technical instruction for the analysis and processing of OCMM data of BIW,but also can provide an example for other fields,such as high-speed train,aerospace,ships etc.