Non-Continuous Industrial Data Analysis Using Discrete Fourier Transform

Analysis of big data is hot topic for exploring the contained values, however, the development of the analytical methods are still a challenging task due to the complexity in structure and variety. In this work , a method for pre-processing and modeling of the non-continuous industrial data was developed and applied in the analysis of a dataset for an industrial production during six years. Four quality parameters and five production parameters were included and the data were collected in batches and sampled in different time and frequency. Fourier transform was used to obtain the frequency composition of the parameters , and then reconstructed data for each parameter were calculated by the inverse transform using the same time schedule. Therefore , the data of all the parameters at the same time points could be obtained and the missing values in the raw data could be filled , making the reconstructed data suitable for building the model between the quality and production parameters. Furthermore, the smoothing effect could be observed in the reconstructed data. Four models were built for the four quality parameters , all of which had a reliable prediction with the mean bias less than 5%.