The Research Of Fault Identification Method In The Procedure Of Nickel Electrolysis Based On Image Feature Analysis

Nickel are indispensable strategic metals in China.In nickel smelting industry,nickel products are mainly produced through electrolytic refining process of nickel sulfide anode.During the procedure of nickel electrolysis,workers need to detect whether the copper rod has abnormal heating conditions or open circuit conditions and check whether the new liquid of the replenishing port is blocked near the cathode copper rod of the electrolytic cell,which is an important part of nickel electrolysis production process management.At present,manual inspection methods are widely used in nickel electrolysis workshop to detect the working conditions of conductive copper rods and new liquids,which makes workload huge and the labor intensity high.Therefore,it is important to realize automatic detection at the nickel electrolysis workshop site and warn the faults in the nickel electrolysis process in time,which can reduce labor costs and labor intensity,and bring economic benefits to nickel production.This paper uses the mature computer vision method to study image processing and recognition on infrared image of copper rods and new liquid image,proposes the fault identification method of copper rod and the new liquid detection method in the procedure of nickel electrolysis.The main content includes the following two parts:1.Fault identification method of conductive copper rod based on infrared image and Support Vector Machine.First,according to the location distribution characteristics and grayscale characteristics of the copper rods in the electrolytic cell,this paper selects different initial seed,and get segmentation result of conductive copper rod by region growing algorithm.Then,analyze and extract the gray feature and Hu moment of copper rod in infrared image to form the feature vector.Finally,use the feature vector and train the Support Vector Machine model to predict the fault types of the copper rod.The model can divide the conductivive states of the copper rod into abnormal heating conditions,normal operating conditions and open circuit conditions,which the accuracy rate of can reach 90%.2.The new liquid detection method based on image processing and recognition.First,for low-light and low-contrast new liquid images,this paper uses the algorithm of adaptive local gamma transformation to achieve image enhancement and improve the contrast of the water flow area.Secondly,it extracts Haar features from the original images of new liquid,and trains the Ada Boost cascade classifier to detect the new liquid near replenishing port.In the end,the model reached 85% accuracy on the existing new liquid sample.At the same time,due to the powerful learning capabilities of the deep learning network,this paper considers using the YOLOv3 deep network model to accomplish new liquid target detection.After training,the YOLOv3 model can reach an accuracy of 92.1%,and has a good ability to recognize difficult samples such as occlusion and reflection.The research work of this paper can accomplish the fault identification task of the nickel electrolysis process well on the existing samples.At same time,based on the existing work,this paper looks forward to adjust and improve the fault identification method when automated detection of mobile platform will realized on the site of nickel electrolysis process.