Study On Signal Identification Of Storage Tank Bottom Corrosion Internal Detection

The number of oil storage tanks has greatly increased with the continuous construction of oil reserve bases, which means higher requirement of safety management. Research has shown that the bottom plate corrosion of large storage tanks is one of the main causes for major accidents. A novel internal method for the tank bottom corrosion detection has been developed in this paper. Superior to the traditional acoustic emission methods, this novel method has made lots of improvement and optimization. An approximating conception was used in the whole detection system. That is the detector going into the storage tank, getting close to the target location and collecting the corrosion signals without mode change. This method can perfectly solve problems such as signal attenuation, relocation and virtual positioning in the traditional acoustic emission methods, thus realizing quantitative measurement. Research in this paper mainly includes the following parts: 1) the introduction of the basic principles; 2) the experimental system compositio n; 3) Setting of the field experiment parameters; 4) the problems appeared during the experimental process and how to solve them with the method of pattern recognition.A data acquisition system for the internal acoustic emission detection was established, including the acoustic emission sensors, the signal conditioning circuit, the data acquisition card and the PC104 control unit. The data acquisition software was programmed by VC++ and the data processing software was programmed by MATLAB. Experiments were conducted both in the laboratory and field. Corrosion signals were collected in the laboratory successfully, while the signals collected in the field test were mixed with lots of back drops of interference.In this study, two novel methods to differentiate the interference from desired corrosion signals in acoustic emission detection process are proposed. These methods are based on extreme learning machine(ELM) and conditional random field(CRF). Firstly, three feature selection and extraction methods: scatter matrix, maximal relevance and minimal redundancy and principal component analysis were applied respectively to optimize initial acoustic emission features. Then ELM and CRF classification models were constructed. The classification results were compared with those of back propagation(BP) neural network, support vector machine(SVM) and hidden Markov model(HMM) according to the accuracy, training time and ROC curve. Finally, b-value method was used to evaluate the classification effect.The experimental results indicate that both ELM and CRF could differentiate the interference from desired corrosion signals. Compared with BP, SVM and HMM, the classification accuracy of CRF is higher than those of others and the training time of ELM is shorter than those of others. Furthermore, the statistical distribution of b-value reflects the process of carbon steel sheet corroded by phosphoric acid. The b-value distribution of corrosion signals identified by ELM and CRF are in accordance with that of corrosion signals in the laboratory.