Research On Comprehensive Fault Diagnosis Technology Of Oxidation Tank Based On Multivariate Data

Oxidation tank is an important equipment for gold extraction by biological oxidation.However,under the influence of high altitude and extreme weather,the oxidation tank operated in Xinjiang for a long time will inevitably have faults of some sensors and actuators,resulting in sharp changes of temperature in the tank.Microorganisms involved in the oxidation reaction will be inactivated or even killed,which will not only affect the gold extraction rate but also cause huge economic losses.So it is necessary to diagnose the oxidation tank working in extreme weather.Through field investigation,various production data of oxidation tank under stable operation and fault state were collected.Combined with the experimental equipment of oxidation tank,the research on comprehensive fault diagnosis technology of oxidation tank based on multivariate data was carried out.The research contents of this paper are as follows:1)Aiming at the goal of determining the fault type and fault characteristics,the internal structure of the oxidation tank was analyzed thermodyna-mically,and several common fault types and measurement indicators in winter in Xinjiang were sorted out.Because there are many measurement indexes and there are linear and nonlinear relations between pairwise,Maximum Information Coefficient-Principal Component Analysis algorithm is introduced to reduce the dimension of related variables,and the complex relationship between variables is transformed to make the newly obtained principal components independent of each other.Finally,the fault features are extracted by calculating the contribution rate to lay a good foundation for the later data analysis.2)Aiming at the target of selecting suitable algorithm for fault diagnosis of oxidation tank,the Expectation Maximization algorithm is used to classify each set of production data,assuming that the sample data distribution conforms to the Gaussian mixture model,initialize model parameters by K-MEANS algorithm,and iteratively update the model parameters until the maximum number of iterations is met.Since the values of the feature amounts are all independent,the obtained model parameters have effects on the classification accuracy,need use the naive Bayes classifier algorithm to construct the probability density function of the characteristic value on a certain fault type,and calculate a more accurate classification result.This paper uses naive Bayes classifier and K-MEANS algorithm to improve the traditional EM algorithm to make up for the shortcomings of the independent model parameters in the traditional EM algorithm to reduce the classification accuracy and slow convergence speed.At the same time,it overcomes the problem that single naive Bayes classifier algorithm needs a lot of data accurate model is overcome.It provides a theoretical framework for the fault diagnosis experiment of oxidation tank in the later stage.3)In order to verify the advancement and superiority of the algorithm used in this paper,heat exchange experiments were performed using the oxidation tank simulation equipment of Siemens laboratory.In combination with the production site,suitable operating data were selected as characteristic quantities.Then mathematically model the sample data,select part of the data for testing to obtain the classification result.Through many experiments,the naive Bayes classifier algorithm and the K-MEANS algorithm are used to improve the EM algorithm,which can not only establish an accurate fault diagnosis model,but also have higher classification accuracy than the current fault diagnosis technology for oxidation tanks.The results of this research provide new directions and new ideas for the fault diagnosis of oxidation tanks in alpine regions.