Research Of Gross Error Detection In Complex Industrial Processes

The purpose of all industrial production is to get qualified products.In order to control the pretty quality,it is necessary to make a real-time and accurate detection for product important process variables which are closely related to the product quality.In industrial controlling processes,many variables need to be measured by using the flow meter or associated sensor.And the collected field data will be as the basis of the following work(such as process controlling,production management,decision analysis and process adjustment).However,there are always gross errors inevitably in the data collected from industrial field because of various reasons(such as the malfunction of measuring instrument,the leakage of piping of equipment and unstable operation).On the one hand,if there are gross errors in the measurement data,the collected data doesn't reflect the real situation of production process properly in the process of solving the optimization problem.And in the data correction process,the gross errors will be allocated to the original correct data so that the collected data will be worse after being corrected than the previous ones.On the other hand,the result data of gross error data will help operator to detect and remove the instrument failure or the pipeline leakage.And according to the above,the operator can maintain instruments and equipment pertinently.Therefore,when corrected and estimated,the data need to be detected whether there are gross errors in order to guarantee the measurement accuracy of soft measurement model.Based on above background,this article expands study for the gross error detection of complex industrial processes.Firstly,the original 3MAD gross error detection algorithm and MMMD clustering algorithm are researched and analyzed.Then we propose a new clustering method based on 3MAD-MMMD which is combined by the two methods together.In addition,in a further study of the new method,we also propose weighted clustering analysis method based on 3MAD-GRW-MMMD.Through comparing these methods,we discovery the advantages of the new one.Secondly,the PCA is carried out relevant research and analysis.Then we propose a reasonable improvement program for the PCA.Through many experiments,we propose a new method which is based on GRW-PCA by comparing at last.Finally,according to the studying basis above,we apply these two new methods to industrial processes,such as the fermentation process of penicillin,LF refining furnace smelting process and the guide disc rev measurement of seamless tube.The experiments and simulations show that 3MAD-GRW-MMMD and GRW-PCA can identify gross errors in the data more quickly and efficiently.In a word,this method is real-time,accurate,economical and efficient and reliable.By comparing and analyzing various characteristics of each industrial process,we finalize to make sure that which method applies to which industrial process.